Introduction¶
This is a notebook created to document a generic Trade Engine for backtesting and paper trading. It includes a review of the architecture, the type of backtests supported, and a library of documented examples to demonstrate the functionality.
This notebook is for demonstration only, the supporting libraries are not yet available for download. The entire environment, database, and this Jupyter Notebook are available via Docker Compose YML file, all that is required for executing is a local Docker environment. The Git Hub Project with instructions to setup and execute this notebook in a docker container are defined at: https://github.com/zambranag777/Trade_Engine_Demo
This notebook, HTML version, and all the backtest examples results were saved to Excel, and can be retrieved from github at:
Zip file with all results save to Excel: https://github.com/zambranag777/Research_Projects/blob/6ad1eb5684cfb9f08150057a6b050239d6456cec/trade_engine_backtest_examples_results_saved_to_excel.zip
Latest version of this Jupyter Notebok: https://github.com/zambranag777/Research_Projects/blob/6ad1eb5684cfb9f08150057a6b050239d6456cec/TradeEngine_Demo.ipynb
Latest version of this Jupyter Notebok in HTML format: https://github.com/zambranag777/Research_Projects/blob/6ad1eb5684cfb9f08150057a6b050239d6456cec/TradeEngine_Demo.html
The Jupyter Notebook should be opened in Jupyter, it will maintain the organization layout, graphics, and table of contents. All backtest results are saved directly on the notebook, but persisted to distinct Excel files for easier review and analysis.
Feedback is appreciated, please email me at: zambranag@yahoo.com
Gustavo
Execution Instructions¶
- Run all cells in Setup
- Run all required imports in Examples/Backtest/Imports
- Monitor System Resources
- Use system specific utilities to monitor the CPU and Memory of the Python process
- If running in a Docker container use: docker stats ( similar to top on Linux)
Setup¶
Python Libraries¶
In [1]:
%load_ext pyinstrument
import os
print( os.environ['PYTHONPATH'] )
%reload_ext autoreload
%autoreload 2
# Python ≥3.6 is required
import sys
assert sys.version_info >= (3, 6)
# Common imports
import numpy as np
import os
import pandas as pd
import yfinance as yf
import feather
import tables
import ta
import socket
from datetime import datetime
# Increaset TF threads for docker containers
import tensorflow as tf
tf.config.threading.set_inter_op_parallelism_threads(44)
tf.config.threading.set_intra_op_parallelism_threads(44)
# Ignore useless warnings (see SciPy issue #5998)
import warnings
warnings.filterwarnings(action="ignore", message="^internal gelsd")
warnings.filterwarnings("ignore", category=RuntimeWarning)
import sqlalchemy
print(sqlalchemy.__version__)
from sqlalchemy.orm import with_polymorphic
# ignore lazy loading warnings
from sqlalchemy import exc as sa_exc
with warnings.catch_warnings():
warnings.simplefilter("ignore", category = sa_exc.SAWarning)
from trade_engine.persist.DBLayer import DBLayer
from trade_engine.persist.DBSchema import DBSchema
Logging¶
In [2]:
import logging
from logging.config import fileConfig
fileConfig('../log/logging.ini')
logging.basicConfig(level=logging.WARN)
logging.getLogger().setLevel(logging.WARN)
Quandal¶
- Quandal requires an account for usage.
- Quandal is own by Nasdaq, URL to sign up for an account: https://data.nasdaq.com/
- Upon signing, a key will be assigned, use that key in the code commented out in the cell below.
- Other options to configure the quandl key in Python: https://github.com/quandl/quandl-python
- There are two examples in this notebook that leverage Quandl for market data:
- a) Cottons Futures - Quandal Market Data - SMA Trade Strategy
- b) Portfolio J
In [3]:
# ***Un-comment the code below, and replace with your key or another configuration method
# *** per the quandl link above
#import quandl
#quandl.ApiConfig.api_key = 'typeyourkeyhere'
Database¶
In [4]:
myDBLayer = DBLayer()
# ***
# *** Connect String for SQL Lite db in memory
# ***
# sqliteConnString = 'sqlite:///:memory:'
# myDBLayer.connect( connString = sqliteConnString )
# ***
# *** Connect String for SQL Lite db file
# ***
# dbTradeEngTestPath = '/Users/gustavozambrana/Google Drive/Python/Projects/Trade_Engine/db/dbTradeEngTest.db'
# ConnString = 'sqlite:///' + dbTradeEngTestPath
# ***
# *** Postgres DB
# ***
ConnString = 'postgresql+pg8000://postgres:postgres@localhost:7777/trade_engine_dev_db'
#ConnString = 'postgresql+pg8000://postgres:postgres@172.17.0.1:7778/trade_engine_demo_db'
# *** Connect to DB, Create DB Schema if not already created, and acquired a DB session
myDBLayer.connect( connString = ConnString )
DBSchema.createDBSchema()
session = DBLayer.getSession()
Optional Drop Table(s)¶
In [4]:
DBSchema.dropAll()
Trade Engine Overview¶
Architecture¶
The main features of the trade engine are:
Performs a backtest for a single trade strategy or a portfolio of trade strategies
Support different kinds of backtests and optimizations of hyper-parameters per trade strategy
Analyzed out-of-sample performance for trade strategies and/or portfolio
Support Technical Analysis Models, as well as AI Based Models to model all trade actions in a trade, that is:
Enter, Exit, Position Sizing, Stop-Loss, Trail-Stop-Loss, etc.
Technical Analysis Models are models that:
- Do not need to be fitted prior to a backtest, such as: SMA, RSI, Bollinger Bands, MACD, etc.
AI Based Models are any models that:
- Required to be fitted with a training and validation date ranges that do not overlapped the backtest period, such as Machine Learning, Deep Learning, and Reinforcement Learning Models.
Support ensemble models for any trade action in a Trade Strategy, i.e.:
- N number of models for entering a trade
- M number of models to exit a trade
- K number of models to decide on stop-loss %
Support mixing Technical Analysis and AI Based Models in ensembles of any trade action
Support persisting all information to execute a backtest in a Relational Database, as well as the results
Setup the architecture to enable distributed cloud execution of backtests
The following are the logical components to define a backtest:
- A **Trade Strategy** is defined via a Trade Asset Register Item, which is an object that contains all components necessary to execute in a backtest
- A **Trade Asset Register Item** has several components:
- Trade Strategy Model
- Manages the actions to perform in a trade strategy (Enter, Exit, Position Size, Stop-Lost, etc.)
- Has a reference to a Market Data Manager Object to get the required data
- Trade Asset Investment Manager
- Manages the funds allocated to the trade strategy
- Broker
- Open, closes, and updates trades (Simulated or Paper-Trading)
- Open, closes, and updates trades (Simulated or Paper-Trading)
- Trade Strategy Model
- A **Trade Strategy Model** has different models controlling the different actions in a trade
- A list of Enter Action models, and corresponding policy rules for the ensemble (required)
- A list of Exit Action models, and corresponding policy rules for the ensemble (optional)
- List of models for all other actions and their ensemble policies (optional)
- A **Portfolio Register Item** defines a portfolio
- It is an object that contains a list of Trade Asset Register Items and a Portfolio Investment Manager
- The Portfolio Investment Manager, manages the funds for the entire portfolio, and re-balances the
funds for the portfolio, if configured to do so, periodically.
- A **Trade Strategy** is defined via a Trade Asset Register Item, which is an object that contains all components necessary to execute in a backtest
Building and Executing any type of backtest consist of the following steps:
- Define a Market Data Object that has all the securities price data
- Define a Broker Object to execute trades (Simulation, Paper Trading )
- Define the models for the trade strategy
- All models for all trade actions are defined via JSON configuration
- The only requirement is one entry model, all others are optional
- Create a Trade Strategy Register Item or Portfolio Register Item object
- supported via a high level API
- Execute Backtest
- Select the desire backtest type for execution (see next section for supported types)
- Analysis of the Results
- Each backtest object provides methods to output metadata, performance metrics, trades detail data, trades summary data, etc. (see Data section for a definition of all data created, and explanation of each parameter output)
- Generate Reports
- Full teardown sheets are supported for each trade strategy or portfolio, as well as comparison against a benchmark (see Reports section under Examples)
Backtests¶
There are different types of backtest supported, the following is a list of the different backtests types and their objectives:
**1. Backtest**
- Executes a Trade Register Item or Portfolio Register Item in a test period
- If any of the Trade Register Items contains at least one trade-action rule that is modeled via an AI Model, the modeled Fit date ranges must be specified (Training and Validation )
- A Backtest generates a large set of performance statistics to analyze the trade strategy, and a set of trade performance plots (supported via the package: https://github.com/ranaroussi/quantstats )
**2. WFA Backtest**
- Re-fit all AI models that a given trade strategy or portfolio depends during a backtest
- The re-fit periodicity and size are configurable parameters, for ex:
- Re-fit all AI Models every 3 months during a backtest, as the backtest progresses, it uses the updated fitted models for the action rules.
**3. Sliding Window WFA Backtest**
- Executes a set of WFA-Backtests by sliding or extending the Training, Validation and Test Periods through time, and by also varying (optionally) the re-fit frequency per backtest.
- Purpose of these set of backtest is to analyze the out-of-sample performance for trade strategies and/or portfolios through different market regimes.
**4. Grid-Search Backtest**
- Run a set of backtests by varying a list of hyper-parameters for each Trade-Strategy or set of Trade Strategies.
- Similar in spirit to https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.GridSearchCV.html
- The Grid-Search Backtest can be applied to a regular Backtest, a WFA-Backtest, or as an option in a Sliding-Window WFA Backtest; in the latter, the grid-search will be conducted prior to each sliding-window iteration.
- The main purpose of this type of backtest, it is optimizing hyper parameters for one or more trade strategies in a backtest or per sliding-window shifting.
**5. Production-Sim Backtest**
- The objective of this backtest is to allow the selection of a one or more trade strategies over time for trading. The trade strategies can vary over time, and have different hyper-parameters resulting from an optimization via the Grid-Search Backtest. Subsequently to selecting n number of Trade Strategies, evaluate them out-of-sample, and repeat this process many times through time using the Sliding Window WFA Backtest.
- The main idea is that the trade strategies selected for trading can vary over time, as well as the hyper-parameters of the underlying models of the trade strategies, and the re-fit frequency. One must be careful to carry this process and selection process up to certain time, and only apply once out-of-sample, otherwise, one is adding bias to the selection process.
Graphical Representation¶
Data¶
For each trade strategy in a **Backtest or WFA-Backtest** metadata and performance metrics are generated.
Metadata¶
Parameter Description
----------------------- --------------------------------------------------------------------
0 ta_reg_item_name Trade Strategy Name
1 ta_reg_item_id Database ID if persisted
2 is_portfolio Indicates if entry correspond to Portfolio summary or T.S.
3 trade_asset Trade Asset associated with Trade Strategy
4 model_type [ Static or Dynamic ], Static ties T.S. to one Trade Asset
5 order_type Simulation for submitting a trade, valid options:
[Market, Stop-Loss, Trail-Stop-Loss, Stop-Loss & Trail-Stop-Loss ]
6 trade_on_etf If underlying trade corresponds to a leverage ETF for the Trade Asset
7 enter_model_unique_name Unique name for enter action model across all Trade Strategies in
the portfolio. For example:
[ ts_sma_1, ts_sma_2 ] correspond to two different SMA T.S.
in the portfolio, configured with different parameters for
modeling enter signals
8 enter_ensemble_model Yes = Ensemble Enter Model, No = Individual enter model
9 enter_num_models Number of enter models in the trade strategy
10 enter_model Enter action model, that is: SMA, MACD, LSTM, GRU, etc. or
composite if an ensemble enter trade strategy in place
11 enter_model_type Type of model, either TA (Technical Analysis signal based) or
AI (Artificial Intellgence based, either ML or DL)
12 enter_models List of all enter models in the trade strategy, if non-ensemble,
this list has only one item, matching the enter_model param
13 exit_model_unique_name Unique name for exit action model across all Trade Strategies in
the portfolio.
14 exit_ensemble_model Yes = Ensemble Exit Model, No = Individual exit model
15 exit_num_models Number of exit models in the trade strategy
16 exit_model Exit action model, that is: SMA, MACD, LSTM, GRU, etc. or
composite if an ensemble exit trade strategy in place
17 exit_model_type Type of model, either TA (Technical Analysis signal based) or
AI (Artificial Intellgence based, either ML or DL)
18 exit_models List of all exit models in the trade strategy, if non-ensemble,
this list has only one item, matching the exit_model param
19 stop_model_unique_name Unique name for stop action model across all Trade Strategies in
the portfolio. For example:
[ ts_stop_1, ts_stop_2 ] correspond to two stop action models
20 stop_ensemble_model Yes = Ensemble Stop Model, No = Individual stop model
21 stop_num_models Number of stop models in the trade strategy
22 stop_model Stop action model, that is: ATR, AI_ATR, etc. or
composite if an ensemble stop trade strategy in place
23 stop_model_type Type of model, either TA (Technical Analysis signal based) or
AI (Artificial Intellgence based, either ML or DL)
24 stop_models List of all stop models in the trade strategy, if non-ensemble,
this list has only one item, matching the stop_model param
25 test_period_len_days For Trade Strategies with at least one AI action model, the
duration of the testing period measured in days.
This corresponds to the backtest date range.
26 tr_period_len_days For Trade Strategies with at least one AI action model, the
duration of the training period measured in days.
27 val_period_len_days For Trade Strategies with at least one AI action model, the
duration of the validation period measured in days.Performance Metrics¶
- Generated by Python QuantStats package ( https://github.com/ranaroussi/quantstats )
- Refer to their documentation for description
1 index
2 Start Period
3 End Period
4 Risk-Free Rate
5 Time in Market
6 Cumulative Return
7 CAGR%
8 Sharpe
9 Sortino
10 Sortino/√2
11 Max Drawdown
12 Longest DD Days
13 Volatility (ann.)
14 Calmar
15 Skew
16 Kurtosis
17 Expected Daily %
18 Expected Monthly %
19 Expected Yearly %
20 Kelly Criterion
21 Risk of Ruin
22 Daily Value-at-Risk
23 Expected Shortfall (cVaR)
24 Gain/Pain Ratio
25 Gain/Pain (1M)
26 Payoff Ratio
27 Profit Factor
28 Common Sense Ratio
29 CPC Index
30 Tail Ratio
31 Outlier Win Ratio
32 Outlier Loss Ratio
33 MTD
34 3M
35 6M
36 YTD
37 1Y
38 3Y (ann.)
39 5Y (ann.)
40 10Y (ann.)
41 All-time (ann.)
42 Best Day
43 Worst Day
44 Best Month
45 Worst Month
46 Best Year
47 Worst Year
48 Avg. Drawdown
49 Avg. Drawdown Days
50 Recovery Factor
51 Ulcer Index
52 Avg. Up Month
53 Avg. Down Month
54 Win Days %
55 Win Month %
56 Win Quarter %
57 Win Year % Trade Transaction Detail Data¶
Parameter Description
----------------------- -----------------------------------------------------------------
0 ta_reg_item_name
1 ta_reg_item_id
2 tr_num
3 tr_id
4 broker_id
5 trade_asset
6 trade_funds_start
7 trade_funds_end
8 trade_return_end
9 trade_type
10 enter_trade_next_open_period
11 trade_clerance_percent
12 trade_on_etf
13 trade_completed
14 trade_status
15 trade_status_msg
16 order_type
17 trade_entered_executed_price
18 trade_exited_executed_price
19 num_shares
20 trade_entered_date
21 trade_close_date
22 transaction_fee
23 transaction_tax
24 total_transaction_cost
25 stop_loss_per
26 trail_stop_loss_per
27 trade_high_price_at_exit
28 trade_low_price_at_exit
29 trade_high_price_prev_period
30 trade_low_price_prev_period
31 close_stock_price_for_trade_exit_date
32 high_stock_price_for_trade_exit_date
33 low_stock_price_for_trade_exit_date
34 open_stock_price_for_trade_exit_date
35 final_stop_loss_triggered_price
36 final_trail_stop_loss_triggered_price
37 trade_exit_cause Examples¶
Summary¶
This section gives an overview of the current examples provided, and general instructions
The examples are categorized as follows:
- **Backtest**
- **WFA Backtest**
- **Sliding Window WFA Backtest**
- **Grid Backtest**
- **Production Sim Backtest**
- **Reports**
- **Other**
- **Backtest**
There is a common set of imports for all examples at the beginning that is required to be imported
The examples should be reviewed sequentially, as they get more complex and present more features as they progress. In addition, some examples refered to explanations and models from previous ones.
Every example has a detail description of the backtest at the start, the trade strategies in the backtest, the trade actions configured and their supporting models, their Training and Validation Periods (only required if at least one AI model is present in the backtest), Test Period, etc.
Saving and Re-Storing any objects to a database is optional, in some examples is presented, but it is the same procedure for all examples/objects. In all cases, saving an object, cascades and saves all children objects, and retreiving an object, retrieves all children objects (lazy loading). Refer to SQLAlchemy for additional details.
Examples:
**Backtest**
1. Nasdaq ETF - Yahoo Finance Market Data - SMA Trade Strategy
- **Type :** Trade Strategy
- **Enter & Exit Models :** SMA
- **Market Data Source:** Yahoo Finance
- **Description:** One trade strategy: Nasdaq using Simple Moving Average(SMA) for entering/exiting a trade
2. Cottons Futures - Quandal Market Data - SMA Trade Strategy
- **Type :** Trade Strategy
- **Enter & Exit Models :** SMA
- **Market Data Source:** Quandal
- **Description:** One trade strategy: Cotton's Futures using Simple Moving Average(SMA) for entering/exiting a trade
3. Nasdaq ETF - Yahoo Finance Market Data - LSTM Trade Strategy
- **Type :** Trade Strategy
- **Enter & Exit Models :** LSTM
- **Market Data Source:** Yahoo Finance
- **Description:** One trade strategy: Nasdaq ETF using a LSTM-Model for entering/exiting a trade
4. Portfolio A: Buy-and-Hold and SMA Trade Strategies for US Indixes
- **Type :** Portfolio
- **Enter & Exit Models :** SMA
- **Market Data Source:** Yahoo Finance
- **Description:** Portfolio with the following trade strategies:
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold strategy
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones using SMA
5. Portfolio B: Buy-and-Hold and AI LSTM based Strategies for US Indixes
- **Type :** Portfolio
- **Enter & Exit Models :** LSTM
- **Market Data Source:** Yahoo Finance
- **Description:** Portfolio with the following trade strategies:
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold strategy
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones using LSTM AI Model
6. Portfolio C: Buy-and-Hold and Multiple Configured SMA Trade Strategies for US Indixes
- **Type :** Portfolio
- **Enter & Exit Models :** Seven distinct SMA Models
- **Market Data Source:** Yahoo Finance
- **Description:** Portfolio with the following trade strategies:
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold strategy
- Twenty-One trade strategies: Nasdaq, S&P 500, and DOW Jones using different SMA Models
- Three ensemble trade strategies based on all SMA models, one per US Index
7. Portfolio D: Buy-and-Hold and Multiple Configured LSTM Trade Strategies for US Indixes
- **Type :** Portfolio
- **Enter & Exit Models :** Seven distinct LSTM Models
- **Market Data Source:** Yahoo Finance
- **Description:** Portfolio with the following trade strategies:
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold strategy
- Twenty-One trade strategies: Nasdaq, S&P 500, and DOW Jones using different LSTM Models
- Three ensemble trade strategies based on all LSTM models, one per US Index
8. Portfolio E: Buy-and-Hold and Eleven Distinct Technical Analysis Based Trade Strategies for US Indixes
- **Type :** Portfolio
- **Enter & Exit Models :** Elevent distinct Technical-Analysis (TA) Models
- SMA, Bollinger Bands, MACD, RSI, Stochastic-RSI, ADX, KAMA, TSI, ForceIndex, MFI, SMA
- **Market Data Source:** Yahoo Finance
- **Description:** Portfolio with the following trade strategies:
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold strategy
- Thirty-Three trade strategies: Nasdaq, S&P 500, and DOW Jones for each TA Model
- Three ensemble trade strategies based on all TA models, one per US Index
9. Portfolio F: Buy-and-Hold and Multiple Configured AI Modeled Trade Strategies for US Indixes
- **Type :** Portfolio
- **Enter & Exit Models :** Eight distinct AI based models
- GRU, LSTM, Conv1d-GRU, Conv1d-LSTM, Wavenet, Wavenet-GRU, GRU, Wavenet-LSTM
- **Market Data Source:** Yahoo Finance
- **Description:** Portfolio with the following trade strategies:
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold strategy
- Twenty-Four trade strategies: Nasdaq, S&P 500, and DOW Jones for each AI trading model
- Three ensemble trade strategies based on all AI trading models, one per US Index
10. Portfolio G: All Trade Strategies in Portfolio E and F
- **Type :** Portfolio
- **Enter & Exit Models :** Eight distinct AI based models and Eleven TA based models
- GRU, LSTM, Conv1d-GRU, Conv1d-LSTM, Wavenet, Wavenet-GRU, GRU, Wavenet-LSTM
- SMA, Bollinger Bands, MACD, RSI, Stochastic-RSI, ADX, KAMA, TSI, ForceIndex, MFI, SMA
- **Market Data Source:** Yahoo Finance
- **Description:** Portfolio with the following trade strategies:
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold strategy
- Fifty-Seven trade strategies: Nasdaq, S&P 500, and DOW Jones for each AI and TA trading model
- Three ensemble trade strategies based on all AI and TA trading models, one per US Index
11. Portfolio H: Utilized Leveraged ETFs for Trading on The Model Trade-Asset
- **Type :** Portfolio
- This example shows the effect of creating a Trade Strategy with a given model for any trade asset, but actually, trading on an underlying leveraged ETF for that security
- **Enter & Exit Models :** Same as Portfolio F
- **Market Data Source:** Yahoo Finance
- **Description:** Portfolio with the following trade strategies:
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold strategy
- Five Trade Strategies are executed with ETF-on
- Five Trade Strategies are executed with ETF-off to show effect of ETF-on trading
- Three ensemble trade strategies based on all AI and TA trading models, one per US Index, for the ETF-on Trade Strategies
- Three ensemble trade strategies based on all AI and TA trading models, one per US Index, for the non-ETF Trade Strategies
12. Portfolio I: Monthly Rebalance of Funds in a Portfolio
- **Type :** Portfolio
- **Enter & Exit Models :** Same strategies as defined and explained in Portfolio C
- **Market Data Source:** Yahoo Finance
- **Description:**
- Portfolio Investment Manager configured to balance funds monthly, using an equal weights policy
13. Portfolio J: Trade Strategies with different Market Data Sources (Quandal, YahooFinance, User-Defined DataFrame)
- **Type :** Portfolio
- **Enter & Exit Models :**
- Multiple SMA Models, and LSTM, GRU, MACD, and RSI Models defined to use with each trading asset
- **Market Data Source:** Yahoo Finance, Quandal, User-Defined DataFrame
- **Description:**
- This example shows how to configured a portfolio with trade strategies that depend on different market data sources.
- Each trade strategy is associated with one Market Data Source, which may or may not be shared with other trade strategies.
- This example also shows how to use any market data with the engine, provided it is in a Pandas DataFrame (details discussed in the example, regarding the format of the data frame)
14. Portfolio K: Buy-and-Hold and Multiple Configured SMA Trade Strategies for US Indixes, with Stop Models
- **Type :** Portfolio
- This example shows the effect of using a variety of stop models with SMA models to enter/exit in a trade strategy
- **Enter & Exit Models :** Same as Portfolio C
- **Stop Models :**
- Eight different stop-models
- Stop Loss order with specified stop-loss percent limit
- Trail Stop Loss order with specified trail-stop-loss percent limit
- Combine Stop-Loss & Trail-Stop-Loss order with specified percent limits
- Five different Combine Stop-Loss & Trail-Stop-Loss models, configured with dynamic derived stop-loss and trail-stop-loss percent limits using the Average True Range(ATR) algorithm
- Eight different stop-models
- **Market Data Source:** Yahoo Finance
- **Description:** Portfolio with the following trade strategies:
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold strategy
- 219 Trade Strategies, this is the set of permutations of all enter/exit models, stop models, ensemble strategies, and securities in the portfolio
- The backtest is executed for 20 years, 2001-2020
15. Portfolio L: Buy-and-Hold and Multiple Configured TA Trade Strategies for US Indixes, with Stop Models
- **Type :** Portfolio
- This example shows the effect of using a variety of stop models with a variety of Technical Analysis (TA) models to enter/exit in a trade strategy
- **Enter & Exit Models :** Same as Portfolio E
- **Stop Models :**
- Eight different stop-models
- Stop Loss order with specified stop-loss percent limit
- Trail Stop Loss order with specified trail-stop-loss percent limit
- Combine Stop-Loss & Trail-Stop-Loss order with specified percent limits
- Three different Combine Stop-Loss & Trail-Stop-Loss models, configured with dynamic derived stop-loss and trail-stop-loss percent limits using the Average True Range(ATR) algorithm
- Eight different stop-models
- **Market Data Source:** Yahoo Finance
- **Description:** Portfolio with the following trade strategies:
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold strategy
- 219 Trade Strategies, this is the set of permutations of all enter/exit models, stop models, ensemble strategies, and securities in the portfolio
- The backtest is executed for 20 years, 2001-2020
16. Portfolio M: Buy-and-Hold and Multiple Configured AI Trade Strategies for US Indixes, with Stop Models
- **Type :** Portfolio
- This example shows the effect of using a variety of stop models with AI models to enter/exit in a trade strategy
- **Enter & Exit Models :** Same as Portfolio F
- **Stop Models :**
- Four different stop-models
- Combine Stop-Loss & Trail-Stop-Loss order with specified percent limits
- Three different Combine Stop-Loss & Trail-Stop-Loss models, configured with dynamic derived stop-loss and trail-stop-loss percent limits using the Average True Range(ATR) algorithm
- Four different stop-models
- **Market Data Source:** Yahoo Finance
- **Description:** Portfolio with the following trade strategies:
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold strategy
- 138 Trade Strategies, this is the set of permutations of all enter/exit models, stop models, ensemble strategies, and securities in the portfolio
- The backtest is executed for 1 year, 2009
17. Portfolio N: Buy-and-Hold and Multiple Configured AI Trade Strategies for US Indixes, with AI based Stop Models
- **Type :** Portfolio
- This example shows the effect of using a variety of stop models with AI models to enter/exit in a trade strategy
- **Enter & Exit Models :** Same as Portfolio F
- **Stop Models :**
- Five different stop-models
- Different Combine Stop-Loss & Trail-Stop-Loss models, configured with AI derived stop-loss and trail-stop-loss percent limits using LSTM, GRU, etc.
- The AI time series models are utilized to forecast next day True Range for the stock, and use that information to derive the percent stop and trail-stop limits.
- Five different stop-models
- **Market Data Source:** Yahoo Finance
- **Description:** Portfolio with the following trade strategies:
- Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold strategy
- 138 Trade Strategies, this is the set of permutations of all enter/exit models, stop models, ensemble strategies, and securities in the portfolio
- The backtest is executed for 1 year, 2009
**WFA Backtest**
1. WFA-Backtest-1
- **Type :** Portfolio
- **Enter & Exit Models :** Same entry/exit models configured in Portfolio F
- **Market Data Source:** Yahoo Finance
- **Description:**
- Re-fit AI models every four-months during the backtest
- Don't anchored the training-set
2. WFA-Backtest-2
- **Type :** Portfolio
- **Enter & Exit Models :** Subset of the entry/exit models in Portfolio G
- **Market Data Source:** Yahoo Finance
- **Description:**
- Re-fit AI models every six months during the backtest
- Anchored the training-set
3. WFA-Backtest-3
- **Type :** Portfolio
- **Enter & Exit Models :** Same entry/exit models configured in Portfolio G
- **Market Data Source:** Yahoo Finance
- **Description:**
- Re-fit AI models every six months during the backtest
- Training-set is not anchored
**Slide Window WFA Backtest**
1. SW-WFA-Backtest 1
- **Type :** Portfolio
- **Enter & Exit Models :** Same entry/exit models configured in Portfolio F
- **Market Data Source:** Yahoo Finance
- **Description:**
- Backtest 1: Re-fit AI models every three months
- Backtest 2: Re-fit AI models every six months
- Fixed Training and Validation periods
- Training-set is not anchored during a backtest
- 30 Trade Strategies
2. SW-WFA-Backtest 2
- **Type :** Portfolio
- **Enter & Exit Models :** Same entry/exit models configured in Portfolio F
- **Market Data Source:** Yahoo Finance
- **Description:**
- Re-fit AI models every three months within a backtest
- Slide Datasets by one year increments for five consecutive years
- Training-set is not anchored in outer slide window or during a backtest
- 30 Trade Strategies
3. SW-WFA-Backtest 3
- **Type :** Portfolio
- **Enter & Exit Models :** Same entry/exit models configured in Portfolio G
- **Market Data Source:** Yahoo Finance
- **Description:**
- Re-fit AI trading models every 3 months within a backtest and anchored within the backtest
- Fit datasets are anchored and duration increased by one year increments for five
consecutive years
- Every sliding window iteration, the training set duration is increased by 9 months and the validation set by 3 months, the test set is just shifted by one year.
- Training-set is anchored in outer slide window and also anchored in the re-fitting process during a backtest
- Backtest period from 2003 to 2007, sliding window is one-year
- 63 Trade Strategies
4. SW-WFA-Backtest 4
- **Type :** Portfolio
- **Enter & Exit Models :** Portfolio has the same entry/exit models configured in Portfolio G and C
- **Stop Models :**
- Five different stop-models
- One model with fix values for stop-loss and trail-stop loss
- Four ATR models, each configured with different hyper parameters
- Five different stop-models
- **Market Data Source:** Yahoo Finance
- **Description:**
- The total number of permutations of all trade strategies for all securities, enter/exit
models, stop models, and ensemble models is 633 - Re-fit all AI trading models every 4 and 12 months within a backtest and not anchored within the backtest
- Training-set is not anchored in outer slide window and also not anchored in the re-fitting process during a backtest
- Sliding Window Backtest executed from 2001 to 2010, with one year increments for the
sliding window.
- Two backtests are executed per year, the first one refits the AI models in each Trade Strategy every 4 months, the second one only at the start of the backtest
- The total number of permutations of all trade strategies for all securities, enter/exit
5. SW-WFA-Backtest 5
- **Type :** Portfolio
- **Enter & Exit Models :** Portfolio has the same entry/exit models configured in Portfolio C
- **Stop Models :**
- Eight different stop-models
- Stop Loss order with specified stop-loss percent limit
- Trail Stop Loss order with specified trail-stop-loss percent limit
- Combine Stop-Loss & Trail-Stop-Loss order with specified percent limits
- Five different Combine Stop-Loss & Trail-Stop-Loss models, configured with dynamic derived stop-loss and trail-stop-loss percent limits using the Average True Range(ATR) algorithm
- Eight different stop-models
- **Market Data Source:** Yahoo Finance
- **Description:**
- The total number of permutations of all trade strategies for all securities, enter/exit
models, stop models, and ensemble models is 210 - Sliding Window Backtest executed from 2001 to 2020, with one year increments for the sliding window.
- The total number of permutations of all trade strategies for all securities, enter/exit
**Grid Backtest**
To be added
**Production Sim Backtest**
To be added
**Reports**
1. Create Full Tearsheet Reports
- **Description:**
- Create a full tearsheet report for a backtest
- Create a tearsheet report that compares a backtest against a benchmark strategy
2. Create HTML Tearsheet Reports
- **Description:**
- Similar to 1. but saved to an HTML file.
- **Description:**
**Other**
1. Trade Fee Policy
- **Type :** Trade Strategy
- **Models :** SMA
- **Market Data Source:** Yahoo Finance
- **Description:**
- Demonstrate how to apply a flat broker fee per trade transaction or percent based fee per transaction amount
Backtest¶
Imports¶
In [5]:
# *** Imports required for running Trade Strategy examples
from trade_engine.model.TradeStrategyBaseModelPersist import TradeStrategyBaseModelPersist
from trade_engine.model.TradeStrategyModelStatic import TradeStrategyModelStatic
from trade_engine.model.TradeStrategyMarketDataMng import TradeStrategyMarketDataMng
from trade_engine.model.MarketDataPeriod import MarketDataPeriod
from trade_engine.ts_actions.generate.TradeStrategyActionGenerator import TradeStrategyActionGenerator
from trade_engine.ts_actions.generate.TradeAssetRegisterItemGenerator import TradeAssetRegisterItemGenerator
from trade_engine.ts_actions.generate.TradeStrategyConfigData import TradeStrategyConfigData
from trade_engine.investment_manager.TradingAssetInvMngBase import TradingAssetInvMngBase
from trade_engine.registry.TradingAssetExecutionType import TradingAssetExecutionType
from trade_engine.registry.TradingAssetRegisterItem import TradingAssetRegisterItem
from trade_engine.broker.SimBrokerTrading import SimBrokerTrading
from trade_engine.execution.BacktestExecution import BacktestExecution
from trade_engine.dataset_def.DatasetDef import DatasetDef, DatasetType
from trade_engine.dataset_def.FitDatasetsDef import FitDatasetsDef
from trade_engine.reports.BacktestReports import BacktestReports
# *** Additional Imports required for running Portfolio examples
from tradeg.market_data.MarketData import MarketData
from trade_engine.model.TALibPipelineBuilder import TALibPipelineBuilder
from trade_engine.model.TradeStrategyModelPipelineDataMng import TradeStrategyModelPipelineDataMng
from trade_engine.enter_ts.TradeEnterClassicStrategy import TradeEnterClassicStrategy
from trade_engine.exit_ts.TradeExitClassicStrategy import TradeExitClassicStrategy
from trade_engine.model.classic.CacheTradingModel import CacheTradingModel
from trade_engine.model.classic.TradingRulesModel import TradingRulesModel
from trade_engine.model.classic.VotingTradingRulesModel import VotingTradingRulesModel
from trade_engine.exit_ts.TradeExitStack import TradeExitStack
from trade_engine.enter_ts.TradeEnterStack import TradeEnterStack
from trade_engine.investment_manager.PortfolioInvMng import PortfolioInvMng
from trade_engine.investment_manager.PortfolioRebalanceFrequency import PortfolioRebalanceFrequency
from trade_engine.registry.PortfolioRegisterItem import PortfolioRegisterItem
from trade_engine.model.ai.LSTM_Model import LSTM_Model
from trade_engine.model.ai.WaveNet_RNN_AI_TradingModel import WaveNet_RNN_AI_TradingModel
from trade_engine.model.ai.WaveNet_RNN_AI_TradingModel import WaveNet_RNN_AI_Trading_Rules
from trade_engine.model.ai.WaveNet_RNN_AI_PipelineBuilder import WaveNet_RNN_AI_PipelineBuilder
from trade_engine.enter_ts.TradeEnterAIStrategy import TradeEnterAIStrategy
from trade_engine.exit_ts.TradeExitAIStrategy import TradeExitAIStrategy
from trade_engine.ts_actions.policy.TradeEnterSetPolicy import TradeEnterSetPolicy, TradeEnterSetVotingPolicy
from trade_engine.ts_actions.policy.TradeEnterPolicy import TradeEnterPolicy
from trade_engine.ts_actions.policy.TradeExitSetPolicy import TradeExitSetPolicy, TradeExitSetVotingPolicy
from trade_engine.ts_actions.policy.TradeExitPolicy import TradeExitPolicy
# *** Additional Imports required for running WFA Backtest and Slide Win WFA Backtest
from trade_engine.execution.BacktestExecutionWFA import BacktestExecutionWFA, WFASlideWindowPeriod
from trade_engine.execution.BacktestExecutionSlideWindowWFA import BacktestExecutionSlideWindowWFA
from trade_engine.execution.BacktestExecutionSlideWindowWFA import WFASlideWinScenarioFixDatasets
from trade_engine.execution.BacktestExecutionSlideWindowWFA import WFASlideWinScenarioSlidingDatasets
from trade_engine.execution.BacktestExecutionSlideWindowWFA import WFASlideWinScenarioAnchoredSlidingDatasets
# For other examples
from trade_engine.investment_manager.TradeFeePolicy import TradeFeePolicy, TradeFeeType
Trade Strategies¶
Nasdaq ETF - Yahoo Finance Market Data - SMA Trade Strategy¶
SMA Trade Strategy for Nasdaq ETF
Description:
1. $2.0M Budget 2. 100% capital investedTrade Strategies:
1. One trade strategy: Nasdaq using Simple Moving Average(SMA) for entering/exiting a tradeBacktest:
1. Type: regular 2. Test Period: 1. Begin Date = 2001-01-01 2. End Date = 2005-12-31
Trade Asset¶
In [6]:
trade_asset = "^IXIC"
Market Data Manager¶
Market Data Manager configured to use yahoo-finance for obtaining market data
The following TradeStrategyMarketDataMng object will do the following:
1. Download market data for:
[ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
2. Used yahoo-finance as the market provider
3. Save the downloaded data in the directory and file specified by the arguments:
1. dirPathMarketData
2. fileNameMarketData
4. It will request the market data provider to get data from "beg_date_str" to "end_date_str"
5. If end_date_str is not defined (as the case below), all data available from the begin date is
requested.
6. Interval is set to "1d", this parameter is market data provider specific, see Yahoo-Finance,
Quandal for what options they support
7. The **Adj Close** column is configured as the default price column
8. Note that even so leveraged ETFs are requested from 2000-01-01 to the present, the data
actually returned for these stock symbols by Yahoo-Finance doesn't start until 2011.
The behavior is market data provider specificIn [7]:
#dirPathMarketData = '/Users/gustavozambrana/Google Drive/Python/Projects/Trade_Engine/Data/US_Market_Indixes'
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
In [8]:
ts_marketdat_mng.marketDataObj.getDataFrame()
Out[8]:
Simulation Broker¶
In [9]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Model¶
In [10]:
tsConfDat = TradeStrategyConfigData()
# SMA JSON Configuration
#
# Model Configuration:
# win_fast : fast moving average
# win_slow : slow moving average
# Signals to enter and exit a trade based on the fast ma crossing the slow ma
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 20, 'win_slow' : 100 } },
} } )
Trade Strategy, Trade Investment Manager, and Trade Asset Register Item¶
In [11]:
taRegItemGen = TradeAssetRegisterItemGenerator()
ts_registered_obj = taRegItemGen.createSingleTARegItem(
name_ta_reg_item = "Nasdaq ETF Backtest",
# *** Trade Assets & Market Data
trade_asset = trade_asset,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestment = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
)
Display Entry Rules Generated SMA Signals¶
In [12]:
print("Number of Enter Models: ", len(ts_registered_obj.getTradeStrategyModel().getTSEnterSetObj().tsEntries) )
In [13]:
ts_registered_obj.getTradeStrategyModel().getTSEnterSetObj().tsEntries[0].ts_model_pipline_mng.\
getAllModelDataPipelineDataForStockSymbol(stockSymbol = "^IXIC" )
Out[13]:
Execute Backtest¶
In [14]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Nasdaq ETF - SMA Trade Strategy",
timePeriod = DatasetDef(datasetType = DatasetType.Testing,
strBegDate = "2001-01-01",
strEndDate = "2005-12-31" ),
trade_asset_registered_item = ts_registered_obj )
In [15]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [77]:
tsExecuteEngine.getSummaryMetrics().to_excel("Nasdaq ETF SMA Model backtest-metrics.xlsx")
Execution & Investment Metrics¶
In [16]:
tsExecuteEngine.getExecutionMetrics()
Out[16]:
Performance Metrics¶
In [17]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ]
Out[17]:
Persist Backtest to DB (optional )¶
In [18]:
session.add( tsExecuteEngine )
try :
session.commit()
except Exception as inst:
print(type(inst)) # the exception instance
print(inst)
print(inst.args) # arguments stored in .args
session.rollback() # roll back changes
finally:
print(" done")
Cottons Futures - Quandal Market Data - SMA Trade Strategy¶
SMA Trade Strategy for Cottons Future
Description:
1. $2.0M Budget 2. 100% capital investedTrade Strategies:
1. One trade strategy: Cotton's Futures using Simple Moving Average(SMA) for entering/exiting a tradeBacktest:
1. Type: regular 2. Test Period: 1. Begin Date = 2002-01-01 2. End Date = 2005-12-31
Trade Asset¶
In [22]:
trade_asset = "CHRIS/ICE_CT1"
Market Data Manager¶
Trade Assets - Quandal¶
Download the following data from Quandal:
CHRIS/ICE_CT1
Where CHRIS is the Quandal Data Provider, and the name after the / is the Futures symbol.
The format of the JSON to download from Quandal is:
{ quandal-path :
[ description of data, [ columns to extract from the returned Quandal DataFrame ], { "Quandal_Column" : "Rename Quandal Column" } ]}
Note:
1. The nested dictionary to rename the original downloaded Quandal data, it is provided since Quandal does
not have standard output column names.
- Futures data from different data sources might have different returned column names for the same
state, that is "Close" vs. "close", etc.
- For sharing the same market data manager object it is recommended to rename the columns to have
standard names, and match the ones provided by Yahoo Finance
2. Yahoo Finance returns standard equity column names
3. All implemented AI and Technical Models used standard column names (same as Yahoo Finance)
Columns: Adj Close, Close, Open, High, Low, Volume
Index: Datetime
4. In order to use the implemented models the Quandal Market Data or any Custom Market Data must have
similar named columns, or the user can implement his own models and use whatever column names he/she
prefers.
5. In this example, the Quandal data source does not provide all standard column names, only open,
which is re-named to 'Adj Close'. Therefore only some of the built models will be available,
as they required High, Low, such as Bollinger BandsIn [23]:
indexQuantCodeMap = {
"CHRIS/ICE_CT1" : ["Cotton", ["Open", "Volume"], { "Open" : "Adj Close" } ]
}
In [24]:
dirPathMarketDataQuandl = '../Data/TSMarketDat'
beg_date_for_feature_creation = "2000-10-01"
beg_date_str = "2001-01-01"
end_date_str = "2010-12-31"
q_stock_index_list = list ( indexQuantCodeMap.keys() )
ts_marketdat_mng_quandal = TradeStrategyMarketDataMng(\
marketDataProviderName = "quandl",
supportedTradeAssets = q_stock_index_list,
stock_price_col_name = "Adj Close",
fileNameMarketData = "Quandl_Test_1",
dirPathMarketData = dirPathMarketDataQuandl,
beg_date_str = beg_date_for_feature_creation,
end_date_str = end_date_str,
interval = "1d" ,
updateDataPassEndDate = False,
forceCreateMarketData = False,
indexQuantCodeMap = indexQuantCodeMap, # required for quandl market data provider, defined above
# The Quandal provider used does not return Open, High, Low, Close, so map the "Adj Close" column
# to all standard columns in order to utilized some of the models that required that data (ATR, etc.)
create_std_ohlc_cols_dict = { "Open" : "Adj Close", "Low" : "Adj Close",
"High" : "Adj Close", "Close" : "Adj Close" }
)
In [25]:
ts_marketdat_mng_quandal.marketDataObj.getDataFrame()
Out[25]:
Simulation Broker¶
In [26]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Model¶
In [27]:
tsConfDat = TradeStrategyConfigData()
# SMA JSON Configuration
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 20, 'win_slow' : 100 } },
} } )
Trade Strategy, Trade Investment Manager, and Trade Asset Register Item¶
In [28]:
taRegItemGen = TradeAssetRegisterItemGenerator()
ts_registered_obj = taRegItemGen.createSingleTARegItem(
name_ta_reg_item = "Cotton Futures Backtest",
# *** Trade Assets & Market Data
trade_asset = trade_asset,
ts_marketdat_mng = ts_marketdat_mng_quandal,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestment = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
)
Display Entry Rules Generated SMA Signals¶
In [29]:
ts_registered_obj.getTradeStrategyModel().getTSEnterSetObj().tsEntries[0].ts_model_pipline_mng.\
getAllModelDataPipelineDataForStockSymbol(stockSymbol = trade_asset )
Out[29]:
Execute Backtest¶
In [30]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Cottons Futures - SMA Trade Strategy",
timePeriod = DatasetDef(datasetType = DatasetType.Testing,
strBegDate = "2002-01-01",
strEndDate = "2005-12-31" ),
trade_asset_registered_item = ts_registered_obj )
In [31]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [49]:
tsExecuteEngine.getSummaryMetrics().to_excel("Cottons Futures backtest-metrics.xlsx")
Execution & Investment Metrics¶
In [50]:
tsExecuteEngine.getExecutionMetrics()
Out[50]:
Performance Metrics¶
In [51]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ]
Out[51]:
Persist Backtest to DB (optional )¶
In [52]:
session.add( tsExecuteEngine )
try :
session.commit()
except Exception as inst:
print(type(inst)) # the exception instance
print(inst)
print(inst.args) # arguments stored in .args
session.rollback() # roll back changes
finally:
print(" done")
Nasdaq ETF - Yahoo Finance Market Data - LSTM Trade Strategy¶
LSTM Trade Strategy for Nasdaq ETF
Description:
1. $2.0M Budget 2. 100% capital investedTrade Strategies:
1. One trade strategy: Nasdaq using a LSTM-Model for entering/exiting a tradeBacktest:
1. Type: regular 2. AI Models Fit Date Ranges 1. Training Period: 1. Begin Date = 2001-01-01 2. End Date = 2005-12-31 2. Validation Period: 1. Begin Date = 2006-01-01 2. End Date = 2007-12-31 4. Test Period: 1. Begin Date = 2008-01-01 2. End Date = 2008-12-31
Trade Asset¶
In [32]:
trade_asset = "^IXIC"
Market Data Manager¶
Market Data Manager configured to use yahoo-finance for obtaining market data
The following TradeStrategyMarketDataMng object will do the following:
1. Download market data for:
[ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
2. Used yahoo-finance as the market provider
3. Save the downloaded data in the directory and file specified by the arguments:
1. dirPathMarketData
2. fileNameMarketData
4. It will request the market data provider to get data from "beg_date_str" to "end_date_str"
5. If end_date_str is not defined (as the case below), all data available from the begin date is
requested.
6. Interval is set to "1d", this parameter is market data provider specific, see Yahoo-Finance,
Quandal for what options they support
7. The **Adj Close** column is configured as the default price column
8. Note that even so leveraged ETFs are requested from 2000-01-01 to the present, the data
actually returned for these stock symbols by Yahoo-Finance doesn't start until 2011.
The behavior is market data provider specificIn [33]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
In [34]:
ts_marketdat_mng.marketDataObj.getDataFrame()
Out[34]:
Simulation Broker¶
In [56]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Model¶
In [57]:
tsConfDat = TradeStrategyConfigData()
# LSTM JSON Configuration
#
# Model configuration:
# Simple LSTM MOdel with two layers
# n_steps : length of time series
# no_neurons_lay1 : layer 1 neurons
# no_neurons_lay2 : layer 2 neurons
# pred_ahead_periods : Number of prediction points ahead
# epochs : Number of epochs
# ext_trade_dyn : Extend an existent trade dymically based on current period predictions
# re_eval_exit_dyn : Re-evaluate exiting a trade dynamically based on current period predictions
# entry_policy_args :
# lastn_match : Enter a trade if the last n predictions matched
# cross_enter_signal : Enter a trade if the there is a cross-over signal,
# from buy-to-sell, sell-to-buy
# exit_policy_args :
# lastn_match : Exit a trade if the last n predictions matched
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
Trade Strategy, Trade Investment Manager, and Trade Asset Register Item¶
In [58]:
taRegItemGen = TradeAssetRegisterItemGenerator()
ts_registered_obj = taRegItemGen.createSingleTARegItem(
name_ta_reg_item = "Nasdaq ETF -LSTM Model- Backtest",
# *** Trade Assets & Market Data
trade_asset = trade_asset,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestment = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
)
Display Pipeline Data for LSTM Model¶
In [59]:
# Pipeline just normalizes the data for LSTM Model
ts_registered_obj.getTradeStrategyModel().getTSEnterSetObj().tsEntries[0].ts_model_pipline_mng.\
getAllModelDataPipelineDataForStockSymbol(stockSymbol = "^IXIC" )
Out[59]:
Training, Validation and Test Periods for Backtest¶
In Order to backtest a trade strategy or portfolio that has AI models, the following data is required:
Training Date Range
- Date range that will be used to train the AI models in the backtest
Validation Date Range
- Date range that twill be used to validate the AI models
- Test Period Date Range
- This is the out-sample date range, which corresponds to the date range of the backtest
Date Ranges are defined with a DatasetDef object, and have the following parameters:
- strBegDateAIModels : Date to begin training or validation of an AI model
- strBegDate : Date to begin the backtest
- strEndDate : Date to end training, validation and Backtest
Note:
- The param strBegDateAIModels can be defined as the same date as strBegDate, that is the default, if not defined.
- The critical aspect of using this parameter is in the backtest. Some AI Models, such as time series, others using moving averages over many periods, create params with no predicting data for those initial dates. Thus, in order to ensure the backtest starts on the actual date specified, the strBegDateAIModels should be defined as an earlier date, and it should not overlapped with the validation period end-date to avoid any chance of bias by picking at the validation data.
In [60]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "2000-07-01",
strBegDate = "2001-01-01",
strEndDate = "2005-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2005-07-01",
strBegDate = "2006-01-01",
strEndDate = "2007-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2007-07-01",
strBegDate = "2008-01-01",
strEndDate = "2008-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute Backtest¶
In [61]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Nasdaq-ETF LSTM based Trade Strategy",
timePeriod = testing_period,
fitDatasets = fit_datasets,
trade_asset_registered_item = ts_registered_obj )
In [62]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [66]:
tsExecuteEngine.getSummaryMetrics().to_excel("Nasdaq ETF LSTM Model backtest-metrics.xlsx")
Execution & Investment Metrics¶
In [63]:
tsExecuteEngine.getExecutionMetrics()
Out[63]:
Performance Metrics¶
In [64]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ]
Out[64]:
Persist Backtest to DB (optional )¶
In [65]:
session.add( tsExecuteEngine )
try :
session.commit()
except Exception as inst:
print(type(inst)) # the exception instance
print(inst)
print(inst.args) # arguments stored in .args
session.rollback() # roll back changes
finally:
print(" done")
Portfolios¶
Portfolio A¶
Portfolio A: Long and SMA Trade Strategies for US Indixes
Description:
1. Portfolio with 6 trading strategies 2. $2,000,000/strategy 3. No-Rebalance during the backtest 4. 100% capital invested 5. Portfolio investment using equal weights for each trade strategyTrade Strategies:
1. Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold 2. Three trade strategies: Nasdaq, S&P 500, and DOW Jones using SMABacktest:
1. Type: regular 2. Test Period: 1. Begin Date = 2001-01-01 2. End Date = 2001-12-31
Portfolio Trade Assets¶
In [35]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [36]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
In [37]:
ts_marketdat_mng.marketDataObj.getDataFrame().query('Index == "^DJI" ')
Out[37]:
Enter & Exit Models¶
In [10]:
tsConfDat = TradeStrategyConfigData()
# Model Params
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 2 } },
},
} )
Simulation Broker¶
In [11]:
brokerTradingObj = SimBrokerTrading()
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [12]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio A",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
Persist to DB (optional)¶
In [13]:
session.add( portf_registered_obj )
try :
session.commit()
except Exception as inst:
print(type(inst)) # the exception instance
print(inst.args) # arguments stored in .args
session.rollback() # roll back changes
finally:
print(" done")
Re-Store from DB (optional)¶
In [14]:
session.expunge_all()
del portf_registered_obj
for tsObj in session.query(PortfolioRegisterItem):
print(tsObj.name)
if tsObj.name == "Portfolio A" :
break
portf_registered_obj = tsObj
print(portf_registered_obj.id, portf_registered_obj.name)
Execute Backtest¶
In [15]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio A",
timePeriod = DatasetDef(datasetType = DatasetType.Testing,
strBegDate = "2001-01-01",
strEndDate = "2001-12-31" ),
trade_asset_registered_item = portf_registered_obj )
In [16]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [17]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-A backtest-metrics.xlsx")
Execution & Investment Metrics¶
In [18]:
tsExecuteEngine.getExecutionMetrics()
Out[18]:
Performance Metrics¶
In [19]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ]
Out[19]:
Persist Backtest to DB (optional )¶
In [20]:
session.add( tsExecuteEngine )
try :
session.commit()
except Exception as inst:
print(type(inst)) # the exception instance
print(inst)
print(inst.args) # arguments stored in .args
session.rollback() # roll back changes
finally:
print(" done")
Re-Store Backtest¶
In [21]:
session.expunge_all()
del tsExecuteEngine
for tsObj in session.query(BacktestExecution):
print(tsObj.name)
if tsObj.name == "Backtest for SMA based Trade Strategies" :
break
tsExecuteEngine = tsObj
print(tsExecuteEngine.id, tsExecuteEngine.name)
In [22]:
tsExecuteEngine.getExecutionMetrics()
Out[22]:
Portfolio B¶
Portfolio B: Long and AI LSTM based Strategies for US Indixes
Description:
1. Portfolio with 6 trading strategies
2. $2,000,000/strategy
3. No-Rebalance during the backtest
4. 100% capital invested
5. Portfolio investment using equal weights for each trade strategy
Trade Strategies:
1. Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold
2. Three trade strategies: Nasdaq, S&P 500, and DOW Jones using LSTM AI Model
Backtest:
1. Type: regular
2. AI Models Fit Date Ranges
1. Training Period:
1. Begin Date = 2001-01-01
2. End Date = 2005-12-31
2. Validation Period:
1. Begin Date = 2006-01-01
2. End Date = 2007-12-31
4. Test Period:
1. Begin Date = 2008-01-01
2. End Date = 2008-12-31 Portfolio Trade Assets¶
In [4]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [5]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Enter & Exit Models¶
In [6]:
tsConfDat = TradeStrategyConfigData()
# Model Params
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
Simulation Broker¶
In [7]:
brokerTradingObj = SimBrokerTrading()
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [8]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio B",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
Persist to DB (optional)¶
In [28]:
session.add( portf_registered_obj )
try :
session.commit()
except Exception as inst:
print(type(inst)) # the exception instance
print(inst.args) # arguments stored in .args
session.rollback() # roll back changes
finally:
print(" done")
Re-Store from DB (optional)¶
In [29]:
session.expunge_all()
del portf_registered_obj
for tsObj in session.query(PortfolioRegisterItem):
if tsObj.name == "Portfolio B" :
break
portf_registered_obj = tsObj
print(portf_registered_obj.id, portf_registered_obj.name)
Training, Validation and Test Periods for Backtest¶
In [9]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "2000-07-01",
strBegDate = "2001-01-01",
strEndDate = "2005-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2005-07-01",
strBegDate = "2006-01-01",
strEndDate = "2007-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2007-07-01",
strBegDate = "2008-01-01",
strEndDate = "2008-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute Backtest¶
In [10]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio B",
timePeriod = testing_period,
fitDatasets = fit_datasets,
trade_asset_registered_item = portf_registered_obj )
In [11]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [33]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-B backtest-metrics.xlsx")
Execution & Investment Metrics¶
In [34]:
tsExecuteEngine.getExecutionMetrics()
Out[34]:
Performance Metrics¶
In [35]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ]
Out[35]:
Persist Backtest to DB (optional )¶
In [36]:
session.add( tsExecuteEngine )
try :
session.commit()
except Exception as inst:
print(type(inst)) # the exception instance
print(inst.args) # arguments stored in .args
session.rollback() # roll back changes
finally:
print(" done")
Re-Store Backtest¶
In [37]:
session.expunge_all()
del tsExecuteEngine
for tsObj in session.query(BacktestExecution):
if tsObj.name == "Backtest for Portfolio B" :
break
tsExecuteEngine = tsObj
print(tsExecuteEngine.id, tsExecuteEngine.name)
In [38]:
tsExecuteEngine.getExecutionMetrics()
Out[38]:
Portfolio C¶
Portfolio C: Long and Multiple Configured SMA Trade Strategies for US Indixes
Description:
1. Portfolio with 27 trading strategies 2. $2,000,000/strategy 3. No-Rebalance during the backtest 4. 100% capital invested 5. Portfolio investment using equal weights for each trade strategyTrade Strategies:
1. Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold 2. 21 trade strategies: Nasdaq, S&P 500, and DOW Jones using following SMA Configuration: sma-conf fast_mav, slow_mav, enter_if_last_nmatch, enter_if_cross_over_sig, exit_if_last_nmatch 1 20 100 2 True 2 2 10 50 2 True 2 3 10 50 2 True 1 4 15 60 2 True 2 5 15 75 2 True 1 6 7 35 2 True 2 7 7 35 2 True 1 3. Three trade strategies: Nasdaq, S&P 500, and DOW Jones using the following configuration: - A Composite SMA Trade Strategy based on the the first five SMA Configured trade strategies above - The Set Entry cofiguration rules are: a. Majority Voting (default) Three of the Five SMA Indicators configured in the set must signaled the same type of investment decision. b. Enter a trade if the last n voting decisions are the same, set to 1 (default ) c. Enter a trade if there is cross-over signaled after applying rules a) and b), set to False (default) - The Set Exit configuration rules are: a. Majority Voting (Default) - Exit a long trade if the percent of SMA Indicators signaling an exit for a long position is greater that the percent signaling no-exit - Exit a short trade if the percent of SMA Indicators signaling an exit for a short position is greater that the percent signaling no-exit b. Exit a trade if the last n exit decisions are the same, set to 1 (default ) 4. Three trade strategies: Nasdaq, S&P 500, and DOW Jones using the following configuration: - A Composite SMA Trade Strategy based on all the SMA Configured trade strategies - The Set Entry cofiguration rules are: a. Voting Percent The majority of the signals indicating an investment position (either short or long), must also meet a minimum percent of 50% across all the strategies in the set. b. Enter a trade if the last n voting decisions are the same, set to 1 (default ) c. Enter a trade if there is cross-over signaled after applying rules a) and b), set to False (default) - The Set Exit configuration rules are: a. Majority Voting (Default) - Exit a long trade if the percent of SMA Indicators signaling an exit for a long position is greater that the percent signaling no-exit - Exit a short trade if the percent of SMA Indicators signaling an exit for a short position is greater that the percent signaling no-exit b. Exit a trade if the last n exit decisions are the same, set to 1 (default )Backtest:
1. Type: regular 2. Test Period: 1. Begin Date = 2001-01-01 2. End Date = 2001-12-31
Portfolio Trade Assets¶
In [39]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [39]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [41]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [42]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 20, 'win_slow' : 100 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_3",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_4",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 60 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_5",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 75 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_6",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 7, 'win_slow' : 35 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_7",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 7, 'win_slow' : 35 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
Trade Strategies, Trade Inv. Manager, and Trade Asset Regiter Items¶
In [43]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio C",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [44]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Execute Backtest¶
In [45]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio C",
timePeriod = DatasetDef(datasetType = DatasetType.Testing,
strBegDate = "2001-01-01",
strEndDate = "2001-12-31" ),
trade_asset_registered_item = portf_registered_obj )
In [46]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save To Excel¶
In [47]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-C backtest-metrics.xlsx")
Execution & Investment Metrics¶
In [48]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']]
Out[48]:
Performance Metrics¶
In [49]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ]
Out[49]:
Portfolio D¶
Portfolio D: Long and Multiple Configured LSTM Trade Strategies for US Indixes
Description:
1. Portfolio with 27 trading strategies 2. $2,000,000/strategy 3. No-Rebalance during the backtest 4. 100% capital invested 5. Portfolio investment using equal weights for each trade strategyTrade Strategies:
1. Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold 2. 21 trade strategies: Nasdaq, S&P 500, and DOW Jones using following LSTM Configurations: LSTM Models Parameter #1 #2 #3 #4 #5 #6 #7 epochs 20 20 20 20 20 20 20 no neurons layer 1 20 20 20 20 20 20 20 no neurons layer 2 20 20 20 20 20 20 20 time series length 20 100 50 50 40 30 60 no predictions ahead 3 15 5 5 3 3 5 extend trade dyn Yes Yes Yes Yes Yes Yes Yes re_evalaute_exit_dyn Yes Yes Yes Yes Yes Yes Yes enter_if_last_nmatch 1 1 1 1 1 1 1 enter_if_cross_over_sig No No No No No No No exit_if_last_nmatch 1 1 1 2 1 1 1 3. Three trade strategies: Nasdaq, S&P 500, and DOW Jones using the following Composite model configuration: - A Composite LSTM Trade Strategy based on all the LSTM configured trade strategies above - The Set Entry cofiguration rules are: a. Majority Voting (default) b. Enter a trade if the last n voting decisions are the same, set to 1 (default ) c. Enter a trade if there is cross-over signaled after applying rules a) and b), set to False (default) - The Set Exit configuration rules are: a. Majority Voting (Default) - Exit a long trade if the percent of LSTM Indicators signaling an exit for a long position is greater that the percent signaling no-exit - Exit a short trade if the percent of LSTM Indicators signaling an exit for a short position is greater that the percent signaling no-exit b. Exit a trade if the last n exit decisions are the same, set to 1 (default )Backtest:
1. Type: regular 2. AI Models Fit Date Ranges 1. Training Period: 1. Begin Date = 2001-01-01 2. End Date = 2005-12-31 2. Validation Period: 1. Begin Date = 2006-01-01 2. End Date = 2007-12-31 4. Test Period: 1. Begin Date = 2008-01-01 2. End Date = 2008-12-31
Portfolio Trade Assets¶
In [50]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [40]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [52]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [53]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 100, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 15,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm_3",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm_4",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 2 } },
},
} )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm_5",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 40, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm_6",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 30, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm_7",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 60, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 2 } },
},
} )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [54]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio D",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [55]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Training, Validation and Test Periods for Backtest¶
In [56]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "2000-07-01",
strBegDate = "2001-01-01",
strEndDate = "2005-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2005-07-01",
strBegDate = "2006-01-01",
strEndDate = "2007-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2007-07-01",
strBegDate = "2008-01-01",
strEndDate = "2008-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute Backtest¶
In [57]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio D",
timePeriod = testing_period,
fitDatasets = fit_datasets,
trade_asset_registered_item = portf_registered_obj )
In [58]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [59]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-D backtest-metrics.xlsx")
Execution & Investment Metrics¶
In [60]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']]
Out[60]:
Performance Metrics¶
In [61]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ]
Out[61]:
Portfolio E¶
Portfolio E: Long and Eleven Distinct Technical Analysis Based Trade Strategies for US Indixes
Description:
1. Portfolio with 39 trading strategies 2. $2,000,000/strategy 3. No-Rebalance during the backtest 4. 100% capital invested 5. Portfolio investment using equal weights for each trade strategyTrade Strategies:
1. Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold 2. 33 trade strategies: Nasdaq, S&P 500, and DOW Jones using following TA Indicators & Configurations: See: https://technical-analysis-library-in-python.readthedocs.io/en/latest/index.html for explanation on the technical indicator specific parameters TS-No TA-Indicator Parameters 1 sma fast_mav 10 slow_mav 50 enter_if_last_nmatch 2 enter_if_cross_over_sig Yes exit_if_last_nmatch 1 2 bb window 20 window_dev 1 enter_if_last_nmatch 2 enter_if_cross_over_sig Yes exit_if_last_nmatch 1 3 macd window_slow 26 window_fast 12 window_sign 9 enter_if_last_nmatch 2 enter_if_cross_over_sig Yes exit_if_last_nmatch 1 4 rsi enter_long if rsi <= 30 exit_long if rsi > 30 enter_short if rsi >= 70 exit_short if rsi <= 60 enter_if_last_nmatch 2 enter_if_cross_over_sig Yes exit_if_last_nmatch 1 5 srsi window 14 smooth1 3 smooth2 3 enter_long if stochrsi_d >= 0.5 exit_long if stochrsi_d < 0.5 enter_if_last_nmatch 2 enter_if_cross_over_sig Yes exit_if_last_nmatch 1 6 adx window 14 enter_if_last_nmatch 2 enter_if_cross_over_sig Yes exit_if_last_nmatch 1 7 mkama kama_fast_window 10 kama_fast_pow1 2 kama_fast_pow1 30 kama_slow_window 10 kama_slow_pow1 5 kama_slow_pow1 30 enter_if_last_nmatch 2 enter_if_cross_over_sig Yes exit_if_last_nmatch 1 8 tsi window_slow 25 window_fast 5 enter_if_last_nmatch 2 enter_if_cross_over_sig Yes exit_if_last_nmatch 1 9 forceidx window 13 enter_if_last_nmatch 2 enter_if_cross_over_sig Yes exit_if_last_nmatch 1 10 mfi window 14 no mfi points for slope 5 enter_if_last_nmatch 2 enter_if_cross_over_sig Yes exit_if_last_nmatch 1 11 sma fast_mav 15 slow_mav 60 enter_if_last_nmatch 2 enter_if_cross_over_sig Yes exit_if_last_nmatch 1 3. Three trade strategies: Nasdaq, S&P 500, and DOW Jones using the following configuration: - A Composite Trade Strategy based on all the technical-analysis indicators configured above - The Set Entry cofiguration rules are: a. Majority Voting (default) b. Enter a trade if the last n voting decisions are the same, set to 1 (default ) c. Enter a trade if there is cross-over signaled after applying rules a) and b), set to False - The Set Exit configuration rules are: a. Majority Voting (Default) b. Exit a trade if the last n exit decisions are the same, set to 1 (default )Backtest:
1. Type: regular 2. Test Period: 1. Begin Date = 2006-01-01 2. End Date = 2010-12-31
Portfolio Trade Assets¶
In [62]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [42]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [64]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [65]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 60 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_bb",
'actions_conf' : { "entry_exit" :
{ 'model' : "bb",
'type' : "ta",
'obj_args' : { 'window' : 13, 'window_dev' : 1 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_macd",
'actions_conf' : { "entry_exit" :
{ 'model' : "macd",
'type' : "ta",
'obj_args' : { 'window_slow' : 26, 'window_fast' : 12, 'window_sign' : 9,
'enter_long_macd_diff' : 0.5 , 'exi_long_macd_diff' : 0.5,
'enter_short_macd_diff' : -0.5, 'exi_short_macd_diff' : -0.5 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_rsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "rsi",
'type' : "ta",
'obj_args' : { 'rsi_enter_long' : 30.0, 'rsi_exit_long' : 30.0,
'rsi_enter_short' : 70.0, 'rsi_exit_short' : 70.0 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_srsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "srsi",
'type' : "ta",
'obj_args' : { 'window' : 14, 'smooth1' : 3, 'smooth2' : 3,
'srsi_enter_long' : 0.8, 'srsi_exit_long' : 0.8,
'srsi_enter_short' : 0.2, 'srsi_exit_short' : 0.2 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_adx",
'actions_conf' : { "entry_exit" :
{ 'model' : "adx",
'type' : "ta",
'obj_args' : { 'window' : 14, 'adx_enter_long_ge' : 14, 'adx_exit_long_lt' : 14,
'adx_enter_short_ge' : 14, 'adx_exit_short_lt' : 14 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_kama",
'actions_conf' : { "entry_exit" :
{ 'model' : "kama",
'type' : "ta",
'obj_args' : { 'window_fast' : 10, 'pow1_fast' : 2, 'pow2_fast' : 30,
'window_slow' : 10, 'pow1_slow' : 5, 'pow2_slow' : 30 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #9
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_tsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "tsi",
'type' : "ta",
'obj_args' : { 'window_slow' : 25, 'window_fast' : 5 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #10
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_forceindex",
'actions_conf' : { "entry_exit" :
{ 'model' : "forceindex",
'type' : "ta",
'obj_args' : { 'window' : 39 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #11
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_mfi",
'actions_conf' : { "entry_exit" :
{ 'model' : "mfi",
'type' : "ta",
'obj_args' : { 'window' : 14, 'long_slope_win' : 15, 'short_slope_win' : 15 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [66]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio E",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [67]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Execute Backtest¶
In [68]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio E",
timePeriod = DatasetDef(datasetType = DatasetType.Testing,
strBegDate = "2006-01-01",
strEndDate = "2010-12-31" ),
trade_asset_registered_item = portf_registered_obj )
In [69]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [70]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-E backtest-metrics.xlsx")
Execution & Investment Metrics¶
In [71]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']]
Out[71]:
Performance Metrics¶
In [72]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ]
Out[72]:
Portfolio F¶
Portfolio F: Long and Multiple Configured AI-Modeled Trade Strategies for US Indixes
Description:
1. Portfolio with 30 trading strategies 2. $2,000,000/strategy 3. No-Rebalance during the backtest 4. 100% capital invested 5. Portfolio investment using equal weights for each trade strategyTrade Strategies:
1. Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold 2. 24 trade strategies: Nasdaq, S&P 500, and DOW Jones using the following AI-Models and corresponding configurations: Trading AI-Models Parameter #1 #2 #3 #4 #5 #6 #7 ai-model GRU LSTM Conv1d-GRU Conv1d-LSTM Wavenet Wavenet-GRU GRU epochs 20 20 20 20 20 20 20 no neurons layer 1 20 20 20 20 - 20 20 no neurons layer 2 20 20 20 20 - 20 20 time series length 50 50 50 50 50 50 40 no predictions ahead 5 5 5 5 5 5 3 extend trade dyn Yes Yes Yes Yes Yes Yes Yes re_evalaute_exit_dyn Yes Yes Yes Yes Yes Yes Yes enter_if_last_nmatch 1 1 1 1 1 1 1 enter_if_cross_over_sig No No No No No No No exit_if_last_nmatch 1 1 1 1 1 1 1 Parameter #8 ai-model Wavenet-LSTM epochs 20 no neurons layer 1 20 no neurons layer 2 20 time series length 50 no predictions ahead 5 extend trade dyn Yes re_evalaute_exit_dyn Yes enter_if_last_nmatch 1 enter_if_cross_over_sig No exit_if_last_nmatch 1 Note: - the parameters "no neurons layer 1" and "no neurons layer 2" are only applicable to AI Models that have GRU or LSTM layers. - Conv1d is a convulation layer in one dimension (configured with 40 filters ) - Wavenet is an AI Architectured based on a set of Convolutional Layers 3. Three trade strategies: Nasdaq, S&P 500, and DOW Jones using the following Composite model configuration: - A Composite Trade Strategy based on all the AI Trading Models configured above - The Set Entry cofiguration rules are: a. Majority Voting (default) b. Enter a trade if the last n voting decisions are the same, set to 1 (default ) c. Enter a trade if there is cross-over signaled after applying rules a) and b), set to False (default) - The Set Exit configuration rules are: a. Majority Voting (Default) - Exit a long trade if the percent of Trading AI Models signaled an exit for a long position is greater that the percent signaling no-exit - Exit a short trade if the percent of Trading AI Models signaled an exit for a short position is greater that the percent signaling no-exit b. Exit a trade if the last n exit decisions are the same, set to 1 (default )Backtest:
1. Type: regular 2. AI Models Fit Date Ranges 1. Training Period: 1. Begin Date = 2002-01-01 2. End Date = 2006-12-31 2. Validation Period: 1. Begin Date = 2007-01-01 2. End Date = 2008-12-31 3. Test Period: 1. Begin Date = 2009-01-01 2. End Date = 2009-12-31
Portfolio Trade Assets¶
In [73]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [43]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [75]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [76]:
# Define Metadata for several Trading AI-Models for several Trade Strategies in the portfolio and one
# Trade Strategy based on a composite set of all Trading AI-Models. The composite trade strategy
# uses Majority Voting for entering and exiting a trade (default setting)
# The configuraiton data for each Trade Strategy is defined via a JSON/dict, and
# they are added to the TradeStrategyConfigData() object
#
# This configuration method is a short-cut when creating many trade strategies in a portfolio, or
# a composite of many trade strategies. The JSON format follows the general format:
#
# *** Trade Strategy Config:
#
# 1. ts_unique_tag : unique tag to identify a trade strategy in a set
# 2. actions_conf : dict,
# Description: contains the configuration data for each action in a trade strategy, only
# the "entry_exit", sub-tag is required
# Child tags:
# "entry_exit" : required
# "stop_lost" : optional
# "pos_sizing" : optional
# "pyramiding" : optional
# 3. The format of each child-tag/action follows the same parameters as when defined via the class,
# 4. obj_args : custom set of parameters passed to the defined model defined via the model tag, this
# is a dynamic set of parameters and specific to each model, some share common inputs.
#
# Supported built-in models, for the entry_exit action:
# [ "sma", "bb", "macd", "rsi", "srsi", "adx", "kama", "tsi",
# "forceindex", "mfi", 'conv1d_gru', 'conv1d_lstm', 'wavenet',
# 'wavenet_gru', 'wavenet_lstm', 'lstm', 'gru' ]
#
# Custom models can be added via a custom function. The second strategy in the configuration
# below ('ts_unique_tag' : "ts_gru_1") , defines the "custom_create_func" tag and sets equal to
# "create_actions_ai_ts_model", this function should be already defined or imported. Thus for
# trade strategy #2, the user custom function is used. Note, that for trade strategy: ts_gru_2
# the model is also set to "gru", but in this case it is using the built-in gru model.
# The actual implementantions for all supported models are included in this Jupyter Notebook, at
# the beg. of this portfolio example, and a user can copy and altered as they see fit, it would
# affect the built-in functions, which are scope in a pythong package.
#
#
# *** Trade Strategy Set Config:
#
# There is only one set configuration, and it is defined to apply policies on how to treat the
# set actions, this configuration is optional, and defaults for each action are defined below
#
# "entry" : optional
# Defines the policy to drive the trade decision to follow from the set. As explained in earlier
# examples.
# Default policy parameters are:
# Majority Voting,
# incNoInvestInVotingDec = False,
# lastn_match = 1
# cross_over_signal_for_trading
#
# "exit" : optional
# Defines the policy to drive the exit decision to follow from the set. As explained in earlier
# examples.
# Default policy parameters are:
# Majority Voting,
# lastn_match = 1
#
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 40, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [77]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio F",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [78]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Training, Validation and Test Periods for Backtest¶
In [79]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "2001-07-01",
strBegDate = "2002-01-01",
strEndDate = "2006-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2006-07-01",
strBegDate = "2007-01-01",
strEndDate = "2008-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2008-07-01",
strBegDate = "2009-01-01",
strEndDate = "2009-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute Backtest¶
In [80]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio F",
timePeriod = testing_period,
fitDatasets = fit_datasets,
trade_asset_registered_item = portf_registered_obj )
In [81]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [82]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-F backtest-metrics.xlsx")
Execution & Investment Metrics¶
In [83]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']]
Out[83]:
Performance Metrics¶
In [84]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ]
Out[84]:
Portfolio G¶
Portfolio G: Combines all Trade Strategies in Portfolio E and F
Description:
1. Portfolio with 63 trading strategies 2. $2,000,000/strategy 3. No-Rebalance during the backtest 4. 100% capital invested 5. Portfolio investment using equal weights for each trade strategyTrade Strategies:
1. Three trade strategies: Nasdaq, S&P 500, and DOW Jones with buy-and-hold 2. Fifty-Seven trade strategies, 19 trade strategies per US Index, as defined in Portfolio E and F 3. Three trade strategies: Nasdaq, S&P 500, and DOW Jones using the following Composite model configuration: - A Composite Trade Strategy based on all the Trading Models configured above per US Index - The Set Entry cofiguration rules are: a. Majority Voting (default) b. Enter a trade if the last n voting decisions are the same, set to 1 (default ) c. Enter a trade if there is cross-over signaled after applying rules a) and b), set to False (default) - The Set Exit configuration rules are: a. Majority Voting (Default) - Exit a long trade if the percent of Trading Models signaled an exit for a long position is greater that the percent signaling no-exit - Exit a short trade if the percent of Trading Models signaled an exit for a short position is greater that the percent signaling no-exit b. Exit a trade if the last n exit decisions are the same, set to 1 (default )
Backtest:
1. Type: regular
2. AI Models Fit Date Ranges
1. Training Period:
1. Begin Date = 2001-01-01
2. End Date = 2006-12-31
2. Validation Period:
1. Begin Date = 2007-01-01
2. End Date = 2008-12-31
3. Test Period:
1. Begin Date = 2009-01-01
2. End Date = 2009-12-31 Portfolio Trade Assets¶
In [85]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [44]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [87]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [88]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 40, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #9
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #10
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 60 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #11
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_bb",
'actions_conf' : { "entry_exit" :
{ 'model' : "bb",
'type' : "ta",
'obj_args' : { 'window' : 13, 'window_dev' : 1 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #12
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_macd",
'actions_conf' : { "entry_exit" :
{ 'model' : "macd",
'type' : "ta",
'obj_args' : { 'window_slow' : 26, 'window_fast' : 12, 'window_sign' : 9,
'enter_long_macd_diff' : 0.5 , 'exi_long_macd_diff' : 0.5,
'enter_short_macd_diff' : -0.5, 'exi_short_macd_diff' : -0.5 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #13
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_rsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "rsi",
'type' : "ta",
'obj_args' : { 'rsi_enter_long' : 30.0, 'rsi_exit_long' : 30.0,
'rsi_enter_short' : 70.0, 'rsi_exit_short' : 70.0 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #14
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_srsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "srsi",
'type' : "ta",
'obj_args' : { 'window' : 14, 'smooth1' : 3, 'smooth2' : 3,
'srsi_enter_long' : 0.8, 'srsi_exit_long' : 0.8,
'srsi_enter_short' : 0.2, 'srsi_exit_short' : 0.2 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #15
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_adx",
'actions_conf' : { "entry_exit" :
{ 'model' : "adx",
'type' : "ta",
'obj_args' : { 'window' : 14, 'adx_enter_long_ge' : 14, 'adx_exit_long_lt' : 14,
'adx_enter_short_ge' : 14, 'adx_exit_short_lt' : 14 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #16
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_kama",
'actions_conf' : { "entry_exit" :
{ 'model' : "kama",
'type' : "ta",
'obj_args' : { 'window_fast' : 10, 'pow1_fast' : 2, 'pow2_fast' : 30,
'window_slow' : 10, 'pow1_slow' : 5, 'pow2_slow' : 30 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #17
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_tsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "tsi",
'type' : "ta",
'obj_args' : { 'window_slow' : 25, 'window_fast' : 5 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #118
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_forceindex",
'actions_conf' : { "entry_exit" :
{ 'model' : "forceindex",
'type' : "ta",
'obj_args' : { 'window' : 39 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #19
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_mfi",
'actions_conf' : { "entry_exit" :
{ 'model' : "mfi",
'type' : "ta",
'obj_args' : { 'window' : 14, 'long_slope_win' : 15, 'short_slope_win' : 15 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [89]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio G",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [90]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Training, Validation and Test Periods for Backtest¶
In [91]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "2000-07-01",
strBegDate = "2001-01-01",
strEndDate = "2006-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2006-07-01",
strBegDate = "2007-01-01",
strEndDate = "2008-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2008-07-01",
strBegDate = "2009-01-01",
strEndDate = "2009-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute Backtest¶
In [92]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio G",
timePeriod = testing_period,
fitDatasets = fit_datasets,
trade_asset_registered_item = portf_registered_obj )
In [93]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [94]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-G backtest-metrics.xlsx")
Trade Strategies Execution & Investment Metrics¶
DOW Jones
In [95]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("DJI")', engine='python')
Out[95]:
Nasdaq
In [96]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("IXIC")', engine='python')
Out[96]:
S&P 500
In [97]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("GSPC")', engine='python')
Out[97]:
Performance Metrics¶
Dow Jones
In [98]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^DJI" ')
Out[98]:
Nasdaq
In [99]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^IXIC" ')
Out[99]:
S&P 500
In [100]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^GSPC" ')
Out[100]:
Portfolio H¶
Portfolio H: Utilized Leveraged ETFs for Trading on The Model Security
- All models are trained on their corresponding US Index/model, but the trade enter via a broker is with
the underlying leveraged ETF.
- $2,000,000/strategy
- ETF data for US Indixes not available until 2011 in yahoo finance
- Seven Trade Strategies are executed with ETF trading enable
- Additionally, there is a composite trade strategy for all ETF strategies a
Backtest:
1. Type: regular
2. AI Models Fit Date Ranges
1. Training Period:
1. Begin Date = 2007-01-01
2. End Date = 2012-12-31
2. Validation Period:
1. Begin Date = 2013-01-01
2. End Date = 2014-12-31
3. Test Period:
1. Begin Date = 2015-01-01
2. End Date = 2015-12-31 Portfolio Trade Assets¶
In [101]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [45]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
In [103]:
MarketData.getDateRangesForStocks( ts_marketdat_mng.marketDataObj.getDataFrame() )
Out[103]:
Simulation Broker¶
In [104]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [105]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [106]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio H",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1,
# Trade On ETF Params
trade_on_ETFs = True,
trade_assets_etf_map = { "^DJI" : [ "UDOW", "SDOW"],
"^IXIC" : [ "TQQQ", "SQQQ"],
"^GSPC" : [ "SPXL", "SPXS"] },
)
In [107]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Training, Validation and Test Periods for Backtest¶
In [108]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "2006-07-01",
strBegDate = "2007-01-01",
strEndDate = "2012-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2012-07-01",
strBegDate = "2013-01-01",
strEndDate = "2014-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2014-07-01",
strBegDate = "2015-01-01",
strEndDate = "2015-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute Backtest¶
In [109]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio H",
timePeriod = testing_period,
fitDatasets = fit_datasets,
trade_asset_registered_item = portf_registered_obj )
In [110]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [111]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-H backtest-metrics.xlsx")
Trade Strategies Execution & Investment Metrics¶
DOW Jones
In [112]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("DJI")', engine='python')
Out[112]:
Nasdaq
In [113]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("IXIC")', engine='python')
Out[113]:
S&P 500
In [114]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("GSPC")', engine='python')
Out[114]:
Performance Metrics¶
In [115]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^DJI"' )
Out[115]:
In [116]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^IXIC"' )
Out[116]:
In [117]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^GSPC"' )
Out[117]:
Portfolio I¶
Portfolio I: Monthly Rebalance of Funds in the Portfolio
- Same strategies as defined and explained in Portfolio C.
- Only difference is the Portfolio Investment Manager is configured to balance funds monthly,
using an equal weights policy.
Backtest:
1. Type: regular
2. Test Period:
1. Begin Date = 2012-01-01
2. End Date = 2015-12-31 Portfolio Trade Assets¶
In [118]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [46]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [120]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [121]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 20, 'win_slow' : 100 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : True },
'exit_policy_args' : { 'lastn_match' : 2 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : True },
'exit_policy_args' : { 'lastn_match' : 2 } },
},
} )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_3",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : True },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_4",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 60 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : True },
'exit_policy_args' : { 'lastn_match' : 2 } },
},
} )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_5",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 75 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : True },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_6",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 7, 'win_slow' : 35 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : True },
'exit_policy_args' : { 'lastn_match' : 2 } },
},
} )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_7",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 7, 'win_slow' : 35 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : True },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [122]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio I",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Monthly,
portfRebalancePeriods = 1
)
In [123]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Execute Backtest¶
In [124]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio I",
timePeriod = DatasetDef(datasetType = DatasetType.Testing,
strBegDate = "2012-01-01",
strEndDate = "2015-12-31" ),
trade_asset_registered_item = portf_registered_obj )
In [125]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [126]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-I backtest-metrics.xlsx")
Execution & Investment Metrics¶
In [127]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']]
Out[127]:
Performance Metrics¶
In [128]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ]
Out[128]:
Portfolio J¶
Portfolio J: Trade Strategies with different Market Data Sources (Quandal, YahooFinance, User-Defined DataFrame)
- Futures data is sourced from Quandal (Cotton, Soy Beans, and S&P 500 futures)
- US Equity indexes from Yahoo-Finance
- Leverage ETFs TQQQ and SQQQ from the user-defined DataFrame
- Multiple SMA Models, and LSTM, GRU, MACD, and RSI Models defined to use with each trading asset
- Additionally, there is a composite trade strategy for all Futures, and one composite trade
strategy for the US Index equities
- Buy-and-Hold strategies for each trade asset for comparison
- $2,000,000/strategy
Backtest:
1. Type: regular
2. AI Models Fit Date Ranges
1. Training Period:
1. Begin Date = 2002-01-01
2. End Date = 2006-12-31
2. Validation Period:
1. Begin Date = 2007-01-01
2. End Date = 2008-12-31
3. Test Period:
1. Begin Date = 2009-01-01
2. End Date = 2009-12-31 Market Data Manager - Yahoo-Finance¶
In [129]:
portfolio_trade_assets_yf = [ "^DJI", "^IXIC", "^GSPC" ]
In [47]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng_yf = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Market Data Manager - From User-Defined Pandas DataFrame¶
User can provide their own market data via a Pandas DataFrame
Requirements:
1. DataFrame Index : must be called "Date", and be of datetime format
2. Index column : All data associated with a given trade asset. Same exact format as the
yahoofinance above
3. The rest of the columns in the dataframe can be named anything. However, it is recommended
they used standard names as the ones in YahooFinance above; this will make all built-in
AI and Technical Analysis Models available.
4. Note, interval, supportedTradeAssets are constraint and not enforced; the data that is passed
in the dataframe is what is supported.In [131]:
ts_marketdat_mng_df = TradeStrategyMarketDataMng(\
marketDataProviderName = "dataframe",
marketDataDF = ts_marketdat_mng_yf.marketDataObj.getDataFrame(),
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData_DF",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
In [132]:
ts_marketdat_mng_df.marketDataObj.getDataFrame()
Out[132]:
Market Data Manager - Quandal¶
Download the following data from Quandal:
- CHRIS/CME_S2
- CHRIS/ICE_CT1
CHRIS/CME_SP1
Where CHRIS is the Quandal Data Provider, and the name after the / is the Futures symbol.
The format of the JSON to download from Quandal is:
{ quandal-path :
[ description of data, [ columns to extract from the returned Quandal DataFrame ], { "Quandal_Column" : "Rename Quandal Column" } ]}
Note:
1. The nested dictionary to rename the original downloaded Quandal data, it is provided since Quandal does
not have standard output column names.
- Futures data from different data sources might have different returned column names for the same
state, that is "Close" vs. "close", etc.
- For sharing the same market data manager object it is recommended to rename the columns to have
standard names, and match the ones provided by Yahoo Finance
2. Yahoo Finance returns standard equity column names
3. All implemented AI and Technical Models used standard column names (same as Yahoo Finance)
Columns: Adj Close, Close, Open, High, Low, Volume
Index: Datetime
4. In order to use the implemented models the Quandal Market Data or any Custom Market Data must have
similar named columns, or the user can implement his own models and use whatever column names he/she
prefers.
5. In this example, the Quandal data source does not provide all standard column names, only open,
which is re-named to 'Adj Close'. Therefore only some of the built models will be available,
as they required High, Low, such as Bollinger BandsIn [49]:
indexQuantCodeMap = {
"CHRIS/CME_S2" : ["SoyBeans", ["Open", "Volume"], { "Open" : "Adj Close" } ],
"CHRIS/ICE_CT1" : ["Cotton", ["Open", "Volume"], { "Open" : "Adj Close" } ],
"CHRIS/CME_SP1" : ["S&P_500", ["Open", "Volume"], { "Open" : "Adj Close" } ] }
portfolio_trade_assets_quandal = list ( indexQuantCodeMap.keys() )
In [50]:
dirPathMarketDataQuandl = '../Data/TSMarketDat'
beg_date_for_feature_creation = "2000-10-01"
beg_date_str = "2001-01-01"
end_date_str = "2010-12-31"
ts_marketdat_mng_quandal = TradeStrategyMarketDataMng(\
marketDataProviderName = "quandl",
supportedTradeAssets = portfolio_trade_assets_quandal,
stock_price_col_name = "Adj Close",
fileNameMarketData = "Quandl_Test_1_PorfolioRegisterItem_Portf_J",
dirPathMarketData = dirPathMarketDataQuandl,
beg_date_str = beg_date_for_feature_creation,
end_date_str = end_date_str,
interval = "1d" ,
updateDataPassEndDate = False,
forceCreateMarketData = False,
indexQuantCodeMap = indexQuantCodeMap, # required for quandl market data provider, defined above
# The Quandal provider used does not return Open, High, Low, Close, so map the "Adj Close" column
# to all standard columns in order to utilized some of the models that required that data (ATR, etc.)
create_std_ohlc_cols_dict = { "Open" : "Adj Close", "Low" : "Adj Close",
"High" : "Adj Close", "Close" : "Adj Close" }
)
In [51]:
ts_marketdat_mng_quandal.marketDataObj.getDataFrame()
Out[51]:
Simulation Broker¶
In [136]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [137]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 20, 'win_slow' : 100 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_3",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_4",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 60 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_5",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 75 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_6",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 7, 'win_slow' : 35 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_7",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 12, 'win_slow' : 40 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #9
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #10
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_macd",
'actions_conf' : { "entry_exit" :
{ 'model' : "macd",
'type' : "ta",
'obj_args' : { 'window_slow' : 26, 'window_fast' : 12, 'window_sign' : 9,
'enter_long_macd_diff' : 0.5 , 'exi_long_macd_diff' : 0.5,
'enter_short_macd_diff' : -0.5, 'exi_short_macd_diff' : -0.5 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #11
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_rsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "rsi",
'type' : "ta",
'obj_args' : { 'rsi_enter_long' : 30.0, 'rsi_exit_long' : 30.0,
'rsi_enter_short' : 70.0, 'rsi_exit_short' : 70.0 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [138]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItemForMarketDataMngs(
name_ta_reg_item = "Portfolio J",
# *** Trade Assets & Market Data Managers
market_data_mngs = { "yf" : [ portfolio_trade_assets_yf, ts_marketdat_mng_yf,
MarketDataPeriod.Day ],
"quandal" : [ portfolio_trade_assets_quandal, ts_marketdat_mng_quandal,
MarketDataPeriod.Day ],
"custom1" : [ portfolio_trade_assets_yf, ts_marketdat_mng_df,
MarketDataPeriod.Day ],
},
dirPathMarketData = dirPathMarketData,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [139]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Training, Validation and Test Periods for Backtest¶
In [140]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "2001-07-01",
strBegDate = "2002-01-01",
strEndDate = "2006-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2006-07-01",
strBegDate = "2007-01-01",
strEndDate = "2008-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2008-07-01",
strBegDate = "2009-01-01",
strEndDate = "2009-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute Backtest¶
In [141]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio J",
timePeriod = testing_period,
fitDatasets = fit_datasets,
trade_asset_registered_item = portf_registered_obj )
In [142]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [143]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-J backtest-metrics.xlsx")
Execution & Investment Metrics¶
In [144]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("CHRIS")', engine='python')
Out[144]:
Performance Metrics¶
In [145]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'model_type', 'trade_asset', 'Start Period', 'End Period', 'Time in Market ',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ]
Out[145]:
Portfolio K¶
Portfolio K: Long and Multiple Configured SMA Trade Strategies for US Indixes, with Stop Models
Description:
1. Seven distinct SMA Trade Strategies for enter/exit a trade, plus an ensemble trade strategy of all seven SMA models 2. Eight different stop-models 1. Stop Loss order with specified stop-loss percent limit 2. Trail Stop Loss order with specified trail-stop-loss percent limit 3. Combine Stop-Loss & Trail-Stop-Loss order with specified percent limits 4-8. Five different Combine Stop-Loss & Trail-Stop-Loss models, configured with dynamic derived stop-loss and trail-stop-loss percent limits using Average True Range(ATR) algorithm 3. $2,000,000/strategy 3. No-Rebalance during the backtest 4. 100% capital invested 5. Portfolio investment using equal weights for each trade strategyBacktest:
1. Type: regular 2. Test Period: 1. Begin Date = 2001-01-01 2. End Date = 2020-12-31
Portfolio Trade Assets¶
In [52]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [53]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [148]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [149]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 20, 'win_slow' : 100 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_3",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 57 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_4",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 60 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_5",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 75 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_6",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 7, 'win_slow' : 35 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_7",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 12, 'win_slow' : 40 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
Risk-Stop-Loss Models¶
In [150]:
# ***
# *** Configure Several Types of Stop Actions
# ***
# order_type = 1 ( Stop-Loss )
# order_type = 2 ( Trail-Stop-Loss )
# order_type = 3 ( Combine Stop-Loss & Trail-Stop-Loss )
#
# Stop Models :
# fixval { user-defined stop-loss-percent and/or trail-stop-loss-percent }
# atr { uses ATR model to dynamically determined the stop and trail stop loss percents }
# 1) *** Stop-Loss Order Type
tsStopsConfDat_Stop_Loss = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 1 } }
} )
tsStopsConfDat_Stop_Loss.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_fixval",
'actions_conf' : { "stop" :
{ 'model' : "fixval",
'type' : "ta",
'obj_args' : { 'stop_loss_per' : 0.002, 'trail_stop_loss_per' : 0.005 },
},
},
} )
# 2) *** Trail-Stop-Loss Order Type
tsStopsConfDat_Trail_Stop_Loss = TradeStrategyConfigData()
tsStopsConfDat_Trail_Stop_Loss.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 2} }
} )
tsStopsConfDat_Trail_Stop_Loss.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "tsl_fixval",
'actions_conf' : { "stop" :
{ 'model' : "fixval",
'type' : "ta",
'obj_args' : { 'stop_loss_per' : 0.002, 'trail_stop_loss_per' : 0.005 },
},
},
} )
# 3) *** Combine Stop-Loss & Trail-Stop-Loss Order Type - Using Fixed Values
tsStopsConfDat_Stop_Loss_And_Trail_SL_1 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_1.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_1.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_fixval",
'actions_conf' : { "stop" :
{ 'model' : "fixval",
'type' : "ta",
'obj_args' : { 'stop_loss_per' : 0.002, 'trail_stop_loss_per' : 0.005 },
},
},
} )
# 4) *** Combine Stop-Loss & Trail-Stop-Loss Order Type - Using Dynamic Values from ATR
tsStopsConfDat_Stop_Loss_And_Trail_SL_2 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_2.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_2.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_1",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 14,
'stop_loss_mult' : 0.5,
'trail_stop_loss_mult' : 1.0,
'max_stop_loss_per' : 0.05,
'min_stop_loss_per' : 0.002,
'max_trail_stop_loss_per' : 0.05,
'min_trail_stop_loss_per' : 0.005,
},
},
},
} )
# 5) ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_3 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_3.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_3.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_2",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.05,
'trail_stop_loss_mult' : 0.075,
},
},
},
} )
# 6) ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_4 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_4.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_4.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_3",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.01,
'trail_stop_loss_mult' : 0.03,
},
},
},
} )
# 7) ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_5 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_5.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_5.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_4",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.05,
'trail_stop_loss_mult' : 0.075,
'max_stop_loss_per' : 0.005,
'min_stop_loss_per' : 0.002,
'max_trail_stop_loss_per' : 0.005,
'min_trail_stop_loss_per' : 0.002,
},
},
},
} )
# 8) ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_6 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_6.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_6.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_5",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.05,
'trail_stop_loss_mult' : 0.075,
'max_stop_loss_per' : 0.005,
'min_stop_loss_per' : 0.002,
'max_trail_stop_loss_per' : 0.01,
'min_trail_stop_loss_per' : 0.005,
},
},
},
} )
stopsConfDatList = [ tsStopsConfDat_Stop_Loss, tsStopsConfDat_Trail_Stop_Loss,
tsStopsConfDat_Stop_Loss_And_Trail_SL_1, tsStopsConfDat_Stop_Loss_And_Trail_SL_2,
tsStopsConfDat_Stop_Loss_And_Trail_SL_3, tsStopsConfDat_Stop_Loss_And_Trail_SL_4,
tsStopsConfDat_Stop_Loss_And_Trail_SL_5, tsStopsConfDat_Stop_Loss_And_Trail_SL_6
]
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [151]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio K",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
stop_config_models = stopsConfDatList,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [152]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Execute Backtest¶
In [153]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio K",
timePeriod = DatasetDef(datasetType = DatasetType.Testing,
strBegDate = "2001-01-01",
strEndDate = "2020-12-31" ),
trade_asset_registered_item = portf_registered_obj )
In [154]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [155]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-K backtest-metrics.xlsx")
Execution & Investment Metrics¶
DOW
In [156]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("DJI")', engine='python')
Out[156]:
Nasdaq
In [157]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("IXIC")', engine='python')
Out[157]:
S&P 500
In [158]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("GSPC")', engine='python')
Out[158]:
Portfolio L¶
Portfolio L: Long and Multiple Configured TA Trade Strategies for US Indixes, with Stop Models
Description:
1. Eleven distinct TA Trade Strategies for enter/exit a trade, plus an ensemble trade strategy of all eleven TA models 2. Five different stop-models 1. Stop Loss order with specified stop-loss percent limit 2. Trail Stop Loss order with specified trail-stop-loss percent limit 3. Combine Stop-Loss & Trail-Stop-Loss order with specified percent limits 4-5. Three different Combine Stop-Loss & Trail-Stop-Loss models, configured with dynamic derived stop-loss and trail-stop-loss percent limits using Average True Range(ATR) algorithm 3. $2,000,000/strategy 3. No-Rebalance during the backtest 4. 100% capital invested 5. Portfolio investment using equal weights for each trade strategyBacktest:
1. Type: regular 2. Test Period: 1. Begin Date = 2001-01-01 2. End Date = 2020-12-31
Portfolio Trade Assets¶
In [159]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [54]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [161]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [162]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 60 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_bb",
'actions_conf' : { "entry_exit" :
{ 'model' : "bb",
'type' : "ta",
'obj_args' : { 'window' : 13, 'window_dev' : 1 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_macd",
'actions_conf' : { "entry_exit" :
{ 'model' : "macd",
'type' : "ta",
'obj_args' : { 'window_slow' : 26, 'window_fast' : 12, 'window_sign' : 9,
'enter_long_macd_diff' : 0.5 , 'exi_long_macd_diff' : 0.5,
'enter_short_macd_diff' : -0.5, 'exi_short_macd_diff' : -0.5 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_rsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "rsi",
'type' : "ta",
'obj_args' : { 'rsi_enter_long' : 30.0, 'rsi_exit_long' : 30.0,
'rsi_enter_short' : 70.0, 'rsi_exit_short' : 70.0 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_srsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "srsi",
'type' : "ta",
'obj_args' : { 'window' : 14, 'smooth1' : 3, 'smooth2' : 3,
'srsi_enter_long' : 0.8, 'srsi_exit_long' : 0.8,
'srsi_enter_short' : 0.2, 'srsi_exit_short' : 0.2 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_adx",
'actions_conf' : { "entry_exit" :
{ 'model' : "adx",
'type' : "ta",
'obj_args' : { 'window' : 14, 'adx_enter_long_ge' : 14, 'adx_exit_long_lt' : 14,
'adx_enter_short_ge' : 14, 'adx_exit_short_lt' : 14 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_kama",
'actions_conf' : { "entry_exit" :
{ 'model' : "kama",
'type' : "ta",
'obj_args' : { 'window_fast' : 10, 'pow1_fast' : 2, 'pow2_fast' : 30,
'window_slow' : 10, 'pow1_slow' : 5, 'pow2_slow' : 30 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #9
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_tsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "tsi",
'type' : "ta",
'obj_args' : { 'window_slow' : 25, 'window_fast' : 5 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #10
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_forceindex",
'actions_conf' : { "entry_exit" :
{ 'model' : "forceindex",
'type' : "ta",
'obj_args' : { 'window' : 39 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #11
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_mfi",
'actions_conf' : { "entry_exit" :
{ 'model' : "mfi",
'type' : "ta",
'obj_args' : { 'window' : 14, 'long_slope_win' : 15, 'short_slope_win' : 15 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
Risk-Stop-Loss Models¶
In [163]:
# ***
# *** Configure Several Types of Stop Actions
# ***
# order_type = 1 ( Stop-Loss )
# order_type = 2 ( Trail-Stop-Loss )
# order_type = 3 ( Combine Stop-Loss & Trail-Stop-Loss )
#
# Stop Models :
# fixval { user-defined stop-loss-percent and/or trail-stop-loss-percent }
# atr { uses ATR model to dynamically determined the stop and trail stop loss percents }
# *** Stop-Loss Order Type
tsStopsConfDat_Stop_Loss = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 1 } }
} )
tsStopsConfDat_Stop_Loss.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_fixval",
'actions_conf' : { "stop" :
{ 'model' : "fixval",
'type' : "ta",
'obj_args' : { 'stop_loss_per' : 0.002, 'trail_stop_loss_per' : 0.005 },
},
},
} )
# *** Trail-Stop-Loss Order Type
tsStopsConfDat_Trail_Stop_Loss = TradeStrategyConfigData()
tsStopsConfDat_Trail_Stop_Loss.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 2} }
} )
tsStopsConfDat_Trail_Stop_Loss.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "tsl_fixval",
'actions_conf' : { "stop" :
{ 'model' : "fixval",
'type' : "ta",
'obj_args' : { 'stop_loss_per' : 0.002, 'trail_stop_loss_per' : 0.005 },
},
},
} )
# *** Combine Stop-Loss & Trail-Stop-Loss Order Type - Using Fixed Values
tsStopsConfDat_Stop_Loss_And_Trail_SL_1 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_1.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_1.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_fixval",
'actions_conf' : { "stop" :
{ 'model' : "fixval",
'type' : "ta",
'obj_args' : { 'stop_loss_per' : 0.002, 'trail_stop_loss_per' : 0.005 },
},
},
} )
# ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_3 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_3.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_3.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_2",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.05,
'trail_stop_loss_mult' : 0.075,
},
},
},
} )
# ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_4 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_4.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_4.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_3",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.01,
'trail_stop_loss_mult' : 0.03,
},
},
},
} )
stopsConfDatList = [ tsStopsConfDat_Stop_Loss, tsStopsConfDat_Trail_Stop_Loss,
tsStopsConfDat_Stop_Loss_And_Trail_SL_1, tsStopsConfDat_Stop_Loss_And_Trail_SL_3,
tsStopsConfDat_Stop_Loss_And_Trail_SL_4
]
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [164]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio L",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
stop_config_models = stopsConfDatList,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [165]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Execute Backtest¶
In [166]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio L",
timePeriod = DatasetDef(datasetType = DatasetType.Testing,
strBegDate = "2001-01-01",
strEndDate = "2020-12-31" ),
trade_asset_registered_item = portf_registered_obj )
In [167]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [168]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-L backtest-metrics.xlsx")
Execution & Investment Metrics¶
DOW
In [169]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("DJI")', engine='python')
Out[169]:
Nasdaq
In [170]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("IXIC")', engine='python')
Out[170]:
S&P 500
In [171]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("GSPC")', engine='python')
Out[171]:
Portfolio M¶
Portfolio M: Long and Multiple Configured AI Trade Strategies for US Indixes, with Stop Models
Description:
1. Eight distinct AI Trade Strategies for enter/exit a trade, plus an ensemble trade strategy of all AI models 2. Four different stop-models 1. Combine Stop-Loss & Trail-Stop-Loss order with specified percent limits 2-4. Three different Combine Stop-Loss & Trail-Stop-Loss models, configured with dynamic derived stop-loss and trail-stop-loss percent limits using Average True Range(ATR) algorithm 3. $2,000,000/strategy 3. No-Rebalance during the backtest 4. 100% capital invested 5. Portfolio investment using equal weights for each trade strategyBacktest:
1. Type: regular 2. AI Models Fit Date Ranges 1. Training Period: 1. Begin Date = 2002-01-01 2. End Date = 2006-12-31 2. Validation Period: 1. Begin Date = 2007-01-01 2. End Date = 2008-12-31 4. Test Period: 1. Begin Date = 2009-01-01 2. End Date = 2009-12-31
Portfolio Trade Assets¶
In [172]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [55]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [174]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [175]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 40, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
Risk-Stop-Loss Models¶
In [176]:
# ***
# *** Configure Several Types of Stop Actions
# ***
# order_type = 1 ( Stop-Loss )
# order_type = 2 ( Trail-Stop-Loss )
# order_type = 3 ( Combine Stop-Loss & Trail-Stop-Loss )
#
# Stop Models :
# fixval { user-defined stop-loss-percent and/or trail-stop-loss-percent }
# atr { uses ATR model to dynamically determined the stop and trail stop loss percents }
# *** Combine Stop-Loss & Trail-Stop-Loss Order Type - Using Fixed Values
tsStopsConfDat_Stop_Loss_And_Trail_SL_1 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_1.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_1.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_fixval",
'actions_conf' : { "stop" :
{ 'model' : "fixval",
'type' : "ta",
'obj_args' : { 'stop_loss_per' : 0.002, 'trail_stop_loss_per' : 0.005 },
},
},
} )
# *** Combine Stop-Loss & Trail-Stop-Loss Order Type - Using Dynamic Values from ATR
tsStopsConfDat_Stop_Loss_And_Trail_SL_2 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_2.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_2.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_1",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 14,
'stop_loss_mult' : 0.5,
'trail_stop_loss_mult' : 1.0,
'max_stop_loss_per' : 0.05,
'min_stop_loss_per' : 0.002,
'max_trail_stop_loss_per' : 0.05,
'min_trail_stop_loss_per' : 0.005,
},
},
},
} )
# ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_3 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_3.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_3.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_2",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.05,
'trail_stop_loss_mult' : 0.075,
},
},
},
} )
# ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_4 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_4.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_4.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_3",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.01,
'trail_stop_loss_mult' : 0.03,
},
},
},
} )
stopsConfDatList = [ tsStopsConfDat_Stop_Loss_And_Trail_SL_1, tsStopsConfDat_Stop_Loss_And_Trail_SL_2,
tsStopsConfDat_Stop_Loss_And_Trail_SL_3, tsStopsConfDat_Stop_Loss_And_Trail_SL_4,
]
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [177]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio M",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
stop_config_models = stopsConfDatList,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [178]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Execute Backtest¶
In [179]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "2001-07-01",
strBegDate = "2002-01-01",
strEndDate = "2006-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2006-07-01",
strBegDate = "2007-01-01",
strEndDate = "2008-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2008-07-01",
strBegDate = "2009-01-01",
strEndDate = "2009-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
In [180]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio M",
timePeriod = testing_period,
fitDatasets = fit_datasets,
trade_asset_registered_item = portf_registered_obj )
In [181]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
In [182]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-M backtest-metrics.xlsx")
Execution & Investment Metrics¶
DOW
In [183]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("DJI")', engine='python')
Out[183]:
Nasdaq
In [184]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("IXIC")', engine='python')
Out[184]:
S&P 500
In [185]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("GSPC")', engine='python')
Out[185]:
Portfolio N¶
Portfolio N: Long and Multiple Configured AI Trade Strategies for US Indixes, with Stop Models
Description:
1. Eight distinct AI Trade Strategies for enter/exit a trade, plus an ensemble trade strategy of all AI models 2. Five different stop-models Different Combine Stop-Loss & Trail-Stop-Loss models, configured with AI derived stop-loss and trail-stop-loss percent limits using LSTM, GRU, etc. time series models to forecast next day True Range for the stock, and use that information to derive the percent stop limits. The Stop-Models in this Portfolio are AI based, unlike the ones in Portfolio K, which are based on the Technical Indicator, Average True Range (ATR) 3. $2,000,000/strategy 3. No-Rebalance during the backtest 4. 100% capital invested 5. Portfolio investment using equal weights for each trade strategyBacktest:
1. Type: regular 2. AI Models Fit Date Ranges 1. Training Period: 1. Begin Date = 2002-01-01 2. End Date = 2006-12-31 2. Validation Period: 1. Begin Date = 2007-01-01 2. End Date = 2008-12-31 4. Test Period: 1. Begin Date = 2009-01-01 2. End Date = 2009-12-31
Portfolio Trade Assets¶
In [186]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [56]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [188]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [189]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 40, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
Risk-Stop-Loss Models¶
In [190]:
##### ***
# *** Configure Several Types of Stop Actions
# ***
# order_type = 1 ( Stop-Loss )
# order_type = 2 ( Trail-Stop-Loss )
# order_type = 3 ( Combine Stop-Loss & Trail-Stop-Loss )
#
# Stop Models :
# ai_true_range { - uses an AI model to predict the next's day True-Range, the underlying
# predicting AI Model can vary, in this example, two models use LSTM, and
# another one uses xxx.
# - the ai_true_range model applies rules and constraints to the predicted true
# range, and derives the stop_loss and trail_stop_loss percent limits for the trade.
# }
ai_model_x_features = ['Adj Close', 'Volume', 'true_range', 'true_range_vol1' ]
# AI Stop Model #1 (using LSTM for predicting next day True Range)
tsStopsConfDat_Stop_Loss_And_Trail_SL_1 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_1.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_1.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_ai_tr_1",
'actions_conf' : { "stop" :
{ 'model' : "ai_true_range",
'ai_model' : "gru",
'type' : "ai",
'obj_args' : { 'stop_loss_mult' : 0.05,
'trail_stop_loss_mult' : 0.075,
'max_stop_loss_per' : 0.05,
'min_stop_loss_per' : 0.0001,
'max_trail_stop_loss_per' : 0.07,
'min_trail_stop_loss_per' : 0.0001,
},
'ai_obj_args' : { 'n_steps' : 5, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 1,
'epochs' : 20,
'features' : ai_model_x_features,
'features_label' : 'true_range'
}
},
},
} )
# AI Stop Model #2 (using LSTM for predicting next day True Range)
# -> diff, stop loss and trail loss multiplier, slightly smaller than model #1
tsStopsConfDat_Stop_Loss_And_Trail_SL_2 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_2.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_2.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_ai_tr_2",
'actions_conf' : { "stop" :
{ 'model' : "ai_true_range",
'ai_model' : "lstm",
'type' : "ai",
'obj_args' : { 'stop_loss_mult' : 0.01,
'trail_stop_loss_mult' : 0.03,
'max_stop_loss_per' : 0.05,
'min_stop_loss_per' : 0.0001,
'max_trail_stop_loss_per' : 0.07,
'min_trail_stop_loss_per' : 0.0001,
},
'ai_obj_args' : { 'n_steps' : 10, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'epochs' : 20,
'features' : ai_model_x_features,
'features_label' : 'true_range'
}
},
},
} )
# AI Stop Model #3 (using LSTM for predicting next day True Range)
# -> diff, stop loss and trail loss multiplier, bigger than model #1 and #2
tsStopsConfDat_Stop_Loss_And_Trail_SL_3 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_3.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_3.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_ai_tr_3",
'actions_conf' : { "stop" :
{ 'model' : "ai_true_range",
'ai_model' : "lstm",
'type' : "ai",
'obj_args' : { 'stop_loss_mult' : 0.1,
'trail_stop_loss_mult' : 0.125,
'max_stop_loss_per' : 0.05,
'min_stop_loss_per' : 0.0001,
'max_trail_stop_loss_per' : 0.07,
'min_trail_stop_loss_per' : 0.0001,
},
'ai_obj_args' : { 'n_steps' : 10, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'epochs' : 20,
'features' : ai_model_x_features,
'features_label' : 'true_range'
}
},
},
} )
# AI Stop Model #4 (using conv1d_gru for predicting next day True Range)
# -> diff, stop loss and trail loss multiplier, bigger than model #1 and #2
tsStopsConfDat_Stop_Loss_And_Trail_SL_4 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_4.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_4.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_ai_tr_4",
'actions_conf' : { "stop" :
{ 'model' : "ai_true_range",
'ai_model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'stop_loss_mult' : 0.01,
'trail_stop_loss_mult' : 0.03,
'max_stop_loss_per' : 0.05,
'min_stop_loss_per' : 0.0001,
'max_trail_stop_loss_per' : 0.07,
'min_trail_stop_loss_per' : 0.0001,
},
'ai_obj_args' : { 'n_steps' : 20, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'epochs' : 20,
'features' : ai_model_x_features,
'features_label' : 'true_range'
}
},
},
} )
# AI Stop Model #5 (using conv1d_gru for predicting next day True Range)
# -> diff, stop loss and trail loss multiplier, bigger than model #1 and #2
tsStopsConfDat_Stop_Loss_And_Trail_SL_5 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_5.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_5.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_ai_tr_5",
'actions_conf' : { "stop" :
{ 'model' : "ai_true_range",
'ai_model' : "wavenet",
'type' : "ai",
'obj_args' : { 'stop_loss_mult' : 0.03,
'trail_stop_loss_mult' : 0.05,
'max_stop_loss_per' : 0.05,
'min_stop_loss_per' : 0.0001,
'max_trail_stop_loss_per' : 0.07,
'min_trail_stop_loss_per' : 0.0001,
},
'ai_obj_args' : { 'n_steps' : 20, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'epochs' : 20,
'features' : ai_model_x_features,
'features_label' : 'true_range'
}
},
},
} )
stopsConfDatList = [ tsStopsConfDat_Stop_Loss_And_Trail_SL_1, tsStopsConfDat_Stop_Loss_And_Trail_SL_2,
tsStopsConfDat_Stop_Loss_And_Trail_SL_3, tsStopsConfDat_Stop_Loss_And_Trail_SL_4,
tsStopsConfDat_Stop_Loss_And_Trail_SL_5 ]
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [191]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio N",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
stop_config_models = stopsConfDatList,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [192]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Execute Backtest¶
In [193]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "2001-07-01",
strBegDate = "2002-01-01",
strEndDate = "2006-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2006-07-01",
strBegDate = "2007-01-01",
strEndDate = "2008-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2008-07-01",
strBegDate = "2009-01-01",
strEndDate = "2009-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
In [194]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Portfolio N",
timePeriod = testing_period,
fitDatasets = fit_datasets,
trade_asset_registered_item = portf_registered_obj )
In [195]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [196]:
tsExecuteEngine.getSummaryMetrics().to_excel("Portfolio-N backtest-metrics.xlsx")
Execution & Investment Metrics¶
DOW
In [197]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("DJI")', engine='python')
Out[197]:
Nasdaq
In [198]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("IXIC")', engine='python')
Out[198]:
S&P 500
In [199]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("GSPC")', engine='python')
Out[199]:
Trades Detail Metrics¶
In [200]:
tsExecuteEngine.getTradesTransactionDetailsDF().to_excel("portf_N_trades_transaction_details.xlsx")
WFA Backtest¶
A type of backtests that are designed to simulate fitting AI Models to different date ranges or to shift the date ranges for a backtest through time.
- Description :
- Re-fit the AI models that a given trade strategy depends on during a backtest
- Scenario:
- Given the following initial setting for a backtest:
AI Models Fit Date Ranges
1. Training Period:
1. Begin Date = 2001-01-01
2. End Date = 2005-12-31
2. Validation Period:
1. Begin Date = 2006-01-01
2. End Date = 2007-06-30
Test Period:
1. Begin Date = 2008-01-01
2. End Date = 2008-12-31
- Re-fit the models after the backtest has executed for a period (n days, n weeks, n months), by
either shifting the Training and Validation period or extending, that is anchoring the Training
Period start date
Example 1
- shifting Training and Validation periods
- re-fit every three months
- the AI Models will be trained and validate with the following date ranges:
- Iteration 1:
1. Training Period: 1. Begin Date = 2001-01-01 2. End Date = 2005-12-31 2. Validation Period: 1. Begin Date = 2006-01-01 2. End Date = 2007-06-30 - Iteration 2:
1. Training Period: 1. Begin Date = 2001-04-01 2. End Date = 2006-03-31 2. Validation Period: 1. Begin Date = 2006-04-01 2. End Date = 2007-09-30- Iteration n: ...
- Iteration 1:
- The backtest, for dates between 2008-04-01 and 2008-06-30, will use the updated re-fitted models
- The re-fit process will continue throughout the backtest until the backtest ends.
Example 2
- anchored Training Set
- re-fit every three months
- the AI Models will be trained and validate with the following date ranges:
- Iteration 1:
1. Training Period: 1. Begin Date = 2001-01-01 2. End Date = 2005-12-31 2. Validation Period: 1. Begin Date = 2006-01-01 2. End Date = 2007-06-30 - Iteration 2:
1. Training Period: 1. Begin Date = 2001-01-01 2. End Date = 2006-03-31 2. Validation Period: 1. Begin Date = 2006-04-01 2. End Date = 2007-09-30 - Iteration n:
...
- Iteration 1:
- The backtest, for dates between 2008-04-01 and 2008-06-30, will use the updated re-fitted models
- The re-fit process will continue throughout the backtest
WFA-Backtest 1¶
Description:
- Portfolio uses all trade strategies in Portfolio F
- Re-fit AI trading models every 4 months. For this example, there will be a total of three times
the AI models are fitted, that is:
- Before WFA Backtest starts
- May 1
- September 1
- Training-set is not anchored
Backtest:
1. Type: WFA Backtest
2. AI Models Fit Date Ranges
1. Training Period:
1. Begin Date = 2000-01-01
2. End Date = 2005-12-31
2. Validation Period:
1. Begin Date = 2006-01-01
2. End Date = 2007-12-31
4. Test Period:
1. Begin Date = 2008-01-01
2. End Date = 2008-12-31 Portfolio Trade Assets¶
In [18]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [19]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "1980-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData_From_1980",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
In [20]:
MarketData.getDateRangesForStocks( ts_marketdat_mng.marketDataObj.getDataFrame() )
Out[20]:
Simulation Broker¶
In [21]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [67]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 40, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [23]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "WFA-Backtest 1",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [24]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Training, Validation and Test Periods for Backtest¶
In [25]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "1999-07-01",
strBegDate = "2000-01-01",
strEndDate = "2005-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2005-07-01",
strBegDate = "2006-01-01",
strEndDate = "2007-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2007-07-01",
strBegDate = "2008-01-01",
strEndDate = "2008-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute WFA Backtest¶
In [26]:
tsExecuteEngine = BacktestExecutionWFA( \
name = "WFA-Backtest 1",
test_period = testing_period,
fit_datasets_start = fit_datasets,
trade_asset_registered_item = portf_registered_obj,
wfa_slide_win_size = 4,
wfa_slide_win_period = WFASlideWindowPeriod.Months,
anchor_fit_dataset = False )
In [27]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [31]:
tsExecuteEngine.getSummaryMetrics().to_excel("WFA-Backtest-1 backtest-metrics.xlsx")
Out[31]:
Execution & Investment Metrics¶
In [29]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']]
Out[29]:
Performance Metrics¶
In [30]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ]
Out[30]:
WFA-Backtest 2¶
Description:
- Portfolio has a subset of the trade strategies in Portfolio G
- Smaller subset of TA and AI models to reduce execution duration
- Re-fit AI trading models every 6 months
- Training-set is anchored
Backtest:
1. Type: WFA Backtest
2. AI Models Fit Date Ranges
1. Training Period:
1. Begin Date = 2000-01-01
2. End Date = 2005-12-31
2. Validation Period:
1. Begin Date = 2006-01-01
2. End Date = 2007-12-31
4. Test Period:
1. Begin Date = 2008-01-01
2. End Date = 2010-12-31 Portfolio Trade Assets¶
In [32]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [33]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "1980-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData_From_1980",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [34]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [35]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
"stop_lost" :
{ 'model' : "fix_perc",
'type' : "ta",
'obj_args' : { 'long_stop_los_per' : 5, 'short_stop_los_per' : 5 } }
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 40, 'no_neurons_lay1' : 30,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #10
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 60 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #16
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_kama",
'actions_conf' : { "entry_exit" :
{ 'model' : "kama",
'type' : "ta",
'obj_args' : { 'window_fast' : 10, 'pow1_fast' : 2, 'pow2_fast' : 30,
'window_slow' : 10, 'pow1_slow' : 5, 'pow2_slow' : 30 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #17
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_tsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "tsi",
'type' : "ta",
'obj_args' : { 'window_slow' : 25, 'window_fast' : 5 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [36]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "WFA-Backtest 2",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [37]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Training, Validation and Test Periods for Backtest¶
In [39]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "1999-07-01",
strBegDate = "2000-01-01",
strEndDate = "2005-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2005-07-01",
strBegDate = "2006-01-01",
strEndDate = "2007-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2007-07-01",
strBegDate = "2008-01-01",
strEndDate = "2010-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute WFA Backtest¶
In [40]:
tsExecuteEngine = BacktestExecutionWFA( \
name = "WFA-Backtest 2",
test_period = testing_period,
fit_datasets_start = fit_datasets,
trade_asset_registered_item = portf_registered_obj,
wfa_slide_win_size = 6,
wfa_slide_win_period = WFASlideWindowPeriod.Months,
anchor_fit_dataset = True )
In [41]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [42]:
tsExecuteEngine.getSummaryMetrics().to_excel("WFA-Backtest-2 backtest-metrics.xlsx")
Trade Strategies Execution & Investment Metrics¶
DOW Jones
In [43]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("DJI")', engine='python')
Out[43]:
Nasdaq
In [44]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("IXIC")', engine='python')
Out[44]:
S&P 500
In [45]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("GSPC")', engine='python')
Out[45]:
Performance Metrics¶
Dow Jones
In [46]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^DJI" ')
Out[46]:
Nasdaq
In [47]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^IXIC" ')
Out[47]:
S&P 500
In [48]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^GSPC" ')
Out[48]:
WFA-Backtest 3¶
Description:
- Same trade strategies as Portfolio G
- Re-fit AI trading models every 6 months
- Training-set is not anchored
Backtest:
1. Type: WFA Backtest
2. AI Models Fit Date Ranges
1. Training Period:
1. Begin Date = 2003-01-01
2. End Date = 2008-12-31
2. Validation Period:
1. Begin Date = 2009-01-01
2. End Date = 2010-12-31
4. Test Period:
1. Begin Date = 2011-01-01
2. End Date = 2012-12-31 Portfolio Trade Assets¶
In [49]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [50]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "1980-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData_From_1980",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [51]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [53]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
"stop_lost" :
{ 'model' : "fix_perc",
'type' : "ta",
'obj_args' : { 'long_stop_los_per' : 5, 'short_stop_los_per' : 5 } }
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 40, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #9
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #10
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 60 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #11
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_bb",
'actions_conf' : { "entry_exit" :
{ 'model' : "bb",
'type' : "ta",
'obj_args' : { 'window' : 13, 'window_dev' : 1 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #12
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_macd",
'actions_conf' : { "entry_exit" :
{ 'model' : "macd",
'type' : "ta",
'obj_args' : { 'window_slow' : 26, 'window_fast' : 12, 'window_sign' : 9,
'enter_long_macd_diff' : 0.5 , 'exi_long_macd_diff' : 0.5,
'enter_short_macd_diff' : -0.5, 'exi_short_macd_diff' : -0.5 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #13
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_rsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "rsi",
'type' : "ta",
'obj_args' : { 'rsi_enter_long' : 30.0, 'rsi_exit_long' : 30.0,
'rsi_enter_short' : 70.0, 'rsi_exit_short' : 70.0 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #14
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_srsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "srsi",
'type' : "ta",
'obj_args' : { 'window' : 14, 'smooth1' : 3, 'smooth2' : 3,
'srsi_enter_long' : 0.8, 'srsi_exit_long' : 0.8,
'srsi_enter_short' : 0.2, 'srsi_exit_short' : 0.2 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #15
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_adx",
'actions_conf' : { "entry_exit" :
{ 'model' : "adx",
'type' : "ta",
'obj_args' : { 'window' : 14, 'adx_enter_long_ge' : 14, 'adx_exit_long_lt' : 14,
'adx_enter_short_ge' : 14, 'adx_exit_short_lt' : 14 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #16
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_kama",
'actions_conf' : { "entry_exit" :
{ 'model' : "kama",
'type' : "ta",
'obj_args' : { 'window_fast' : 10, 'pow1_fast' : 2, 'pow2_fast' : 30,
'window_slow' : 10, 'pow1_slow' : 5, 'pow2_slow' : 30 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #17
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_tsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "tsi",
'type' : "ta",
'obj_args' : { 'window_slow' : 25, 'window_fast' : 5 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #18
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_forceindex",
'actions_conf' : { "entry_exit" :
{ 'model' : "forceindex",
'type' : "ta",
'obj_args' : { 'window' : 39 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #19
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_mfi",
'actions_conf' : { "entry_exit" :
{ 'model' : "mfi",
'type' : "ta",
'obj_args' : { 'window' : 14, 'long_slope_win' : 15, 'short_slope_win' : 15 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [54]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "WFA-Backtest 3",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [55]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Training, Validation and Test Periods for Backtest¶
In [56]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "2002-07-01",
strBegDate = "2003-01-01",
strEndDate = "2008-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2008-07-01",
strBegDate = "2009-01-01",
strEndDate = "2010-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2010-07-01",
strBegDate = "2011-01-01",
strEndDate = "2012-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute WFA Backtest¶
In [57]:
tsExecuteEngine = BacktestExecutionWFA( \
name = "WFA-Backtest 3",
test_period = testing_period,
fit_datasets_start = fit_datasets,
trade_asset_registered_item = portf_registered_obj,
wfa_slide_win_size = 6,
wfa_slide_win_period = WFASlideWindowPeriod.Months,
anchor_fit_dataset = False )
In [58]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [59]:
tsExecuteEngine.getSummaryMetrics().to_excel("WFA-Backtest-3 backtest-metrics.xlsx")
Trade Strategies Execution & Investment Metrics¶
DOW Jones
In [60]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("DJI")', engine='python')
Out[60]:
Nasdaq
In [61]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("IXIC")', engine='python')
Out[61]:
S&P 500
In [62]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("GSPC")', engine='python')
Out[62]:
Performance Metrics¶
Dow Jones
In [ ]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^DJI" ')
Nasdaq
In [ ]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^IXIC" ')
S&P 500
In [ ]:
tsExecuteEngine.getSummaryMetrics()\
[ ['ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^GSPC" ')
Persist Backtest to DB (optional )¶
In [ ]:
session.add( tsExecuteEngine )
try :
session.commit()
except Exception as inst:
print(type(inst)) # the exception instance
print(inst)
print(inst.args) # arguments stored in .args
session.rollback() # roll back changes
finally:
print(" done")
Slide Window WFA Backtest¶
- Description :
- Run a set of backtests by shifting the Training, Validation and Test Periods through time, or
increasing the duration of the Training and Validation Periods through time and shifting only
the Test Period
- Purpose of these set of backtest is to analysis the out-of-sample performance for trade strategies
and/or portfolios through time
- The sliding-window of the training, validation, and test periods are configurable.
- This type of backtest is combined with the regular WFA backtest, which allows to re-fit
models during each sliding window shift per the configuration of the regular WFA Backtest.
- This type of backtest can be thought of as double sliding window, one sliding window for shifting or
increasing the size of the training and validation periods, the second one to control how often
to re-fit models during a backtest.
- Due to the potential large number of backtests that can be executed, only the metadata, and summary
statistics and performance data are stored per backtest.
- All backtest statistical and metadata summary information is stored in a dataframe
- The metadata information stored includes actual Fit Data Ranges, Test Period, and re-fit
frequency during a backtest
- Example A:
- Scenario 1:
- Fix Fit and Test Periods.
- Given the following initial setting for a backtest:
Fit Dataset:
train: 1/1/2000 to 12/31/2004
validation: 1/1/2005 to 12/31/2005
Test Period: 1/1/2006 to 12/31/2006
- Run WFA backtests with the datasets above for different re-fit set frequencies, non-anchored:
- 2 weeks
- 1 month
- 3 month
- This example will run the WFA defined part 1, three-times, re-fitting the AI models each time
per the frequency defined.
- Example B:
- Scenario 2:
- WFA Scenario implementation that slides through time the Fit datasets as well as the test period.
- Given the following initial setting for a backtest:
Fit Dataset:
train: 1/1/2000 to 12/31/2004
validation: 1/1/2005 to 12/31/2005
Test Period: 1/1/2006 to 12/31/2006
- Run WFA backtests from the starting datasets above, incrementing sliding window of the datasets
for the Fit and Test periods by one-year, while the end date for the test period is less than
or equal to 12/31/2020
- Re-fit the AI Models in a backtest every 6 months, don't anchor the training set
- Each Iteration is one complete backtest, and in each backtest the AI Models are fitted at
the start and at July 1 of the current backtest year.
Iteration 1:
train: 1/1/2000 to 12/31/2004
validation: 1/1/2005 to 12/31/2005
Test Period: 1/1/2006 to 12/31/2006
Iteration 2:
train: 1/1/2001 to 12/31/2005
validation: 1/1/2006 to 12/31/2006
Test Period: 1/1/2007 to 12/31/2007
..
..
Iteration Last:
train: 1/1/2014 to 12/31/2018
validation: 1/1/2019 to 12/31/2019
Test Period: 1/1/2020 to 12/31/2020
- Example C:
- Scenario 3:
- WFA Scenario implementation that increases through time the date range of the Fit datasets,
but not the test period, as the fit dataset is anchored to its initial position.
- The train set is anchored, and the train and validation set are expanded in every iteration,
pushing the test period dataset forward in time, but with the same length.
- The increased per iteration of the train and validation set are configurable, as well as
the maximum end date for the test period.
- Given the following initial setting for a backtest:
Fit Dataset:
train: 1/1/2000 to 12/31/2004
validation: 1/1/2005 to 12/31/2005
Test Period: 1/1/2006 to 12/31/2006
- Run WFA backtests from the starting datasets above, incrementing sliding window of the datasets
for the Fit and Test periods by one-year, while the end date for the test period is less than
or equal to 12/31/2020
- Re-fit the AI Models in a backtest every 6 months, don't anchor the training set
- This class will iterate through time by creating updated Fit datsets, and Test Periods, with
one year increments overall to the train and validation period. Train dataset increased by 9 months
per iteration, and Validation dataset by 3 months, the test period is just shifted by the one-year.
- that is:
* Iteration 1:
train: 1/1/2000 to 12/31/2004
validation: 1/1/2005 to 12/31/2005
Test Period: 1/1/2006 to 12/31/2006
* Iteration 2:
train: 1/1/2000 to 9/30/2005 (date range increased by nine-months)
validation: 10/1/2005 to 12/31/2006 (date range increaed by three-months and
shifted by nine-months)
Test Period: 1/1/2007 to 12/31/2007 (date range shifted by one-year)
* Iteration 3:
train: 1/1/2000 to 6/30/2006 (date range increased by nine-months)
validation: 7/1/2006 to 12/31/2007 (date range increaed by three-months and
shifted by nine-months)
Test Period: 1/1/2008 to 12/31/2008 (date range shifted by one-year)
..
..
* Iteration Last:
train: 1/1/2000 to xxx
validation: yyy to 12/31/2019
Test Period: 1/1/2020 to 12/31/2020 SW-WFA-Backtest 1¶
Description:
- Same trade strategies as Portfolio F
- Re-fit AI trading models every 3 months, then every 6 months within a backtest
- Fixed Training and Validation periods
- Training-set is not anchored during a backtest
- $200K/strategy
Backtest:
1. Type: Slide Window WFA Backtest
2. AI Models Fit Date Ranges
1. Training Period:
1. Begin Date = 2000-01-01
2. End Date = 2005-12-31
2. Validation Period:
1. Begin Date = 2006-01-01
2. End Date = 2007-12-31
3. Test Period:
1. Begin Date = 2008-01-01
2. End Date = 2008-12-31 Portfolio Trade Assets¶
In [7]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [8]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "1980-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData_From_1980",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
In [9]:
MarketData.getDateRangesForStocks( ts_marketdat_mng.marketDataObj.getDataFrame() )
Out[9]:
Simulation Broker¶
In [10]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [11]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 40, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [12]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "SW-WFA-Backtest 1",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [13]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Training, Validation and Test Periods for Backtest¶
In [14]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "1999-07-01",
strBegDate = "2000-01-01",
strEndDate = "2005-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2005-07-01",
strBegDate = "2006-01-01",
strEndDate = "2007-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2007-07-01",
strBegDate = "2008-01-01",
strEndDate = "2008-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute Slide Window WFA Backtest¶
In [15]:
tsExecuteEngine = BacktestExecutionSlideWindowWFA( \
name = "SW-WFA-Backtest 1",
trade_asset_registered_item = portf_registered_obj,
wfa_slide_win_size_list = [ 3, 6] , # list of int, associated with wfa_slide_win_period_list
wfa_slide_win_period_list = [ WFASlideWindowPeriod.Months, WFASlideWindowPeriod.Months ],
wfa_sw_scenario_obj = WFASlideWinScenarioFixDatasets(fit_dataset = fit_datasets,
test_period = testing_period ),
anchor_fit_dataset = False )
In [16]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [17]:
tsExecuteEngine.getSummaryMetrics().to_excel("SW-WFA-Backtest-1 backtest-metrics.xlsx")
Execution & Investment Metrics¶
In [ ]:
tsExecuteEngine.getSummaryMetrics().info()
Performance Metrics¶
In [81]:
tsExecuteEngine.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^DJI" ')
Out[81]:
In [82]:
tsExecuteEngine.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^IXIC" ')
Out[82]:
In [83]:
tsExecuteEngine.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^GSPC" ')
Out[83]:
Persist Backtest to DB (optional )¶
In [84]:
session.add( tsExecuteEngine )
try :
session.commit()
except Exception as inst:
print(type(inst)) # the exception instance
print(inst)
print(inst.args) # arguments stored in .args
session.rollback() # roll back changes
finally:
print(" done")
SW-WFA-Backtest 2¶
Description:
- Training-set is not anchored in outer slide window or during a backtest
- Same trade strategies as Portfolio F
- Re-fit AI trading models every 3 months within a backtest
- Fixed Training and Validation periods
- Training-set is not anchored in outer slide window or during a backtest
- $200K/strategy
Backtest:
1. Type: Slide Window WFA Backtest
2. Start AI Models Fit Date Ranges
1. Training Period:
1. Begin Date = 2000-01-01
2. End Date = 2005-12-31
2. Validation Period:
1. Begin Date = 2006-01-01
2. End Date = 2007-12-31
3. Final AI Models Fit Date Ranges
1. Training Period:
1. Begin Date = 2004-01-01
2. End Date = 2009-12-31
2. Validation Period:
1. Begin Date = 2010-01-01
2. End Date = 2011-12-31
4. Start Test Period:
1. Begin Date = 2008-01-01
2. End Date = 2008-12-31
5. Final Test Period:
1. Begin Date = 2012-01-01
2. End Date = 2012-12-31 Portfolio Trade Assets¶
In [21]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [22]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "1980-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData_From_1980",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [23]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [24]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 40, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [25]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "SW-WFA-Backtest 2",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [26]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Training, Validation and Test Periods for Backtest¶
In [27]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "1999-07-01",
strBegDate = "2000-01-01",
strEndDate = "2005-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2005-07-01",
strBegDate = "2006-01-01",
strEndDate = "2007-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2007-07-01",
strBegDate = "2008-01-01",
strEndDate = "2008-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute Slide Window WFA Backtest¶
In [28]:
tsExecuteEngine = BacktestExecutionSlideWindowWFA( \
name = "SW-WFA-Backtest 2",
trade_asset_registered_item = portf_registered_obj,
wfa_slide_win_size_list = [ 3 ] , # list of int, associated with wfa_slide_win_period_list
wfa_slide_win_period_list = [ WFASlideWindowPeriod.Months ],
wfa_sw_scenario_obj = WFASlideWinScenarioSlidingDatasets(\
fit_dataset = fit_datasets,
test_period = testing_period,
slide_win_size = 1,
slide_win_period = WFASlideWindowPeriod.Years,
end_date_last_test_period = datetime(year=2012, month=12, day=31) ),
anchor_fit_dataset = False )
In [29]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [30]:
tsExecuteEngine.getSummaryMetrics().to_excel("SW-WFA-Backtest-2 backtest-metrics.xlsx")
Performance Metrics¶
In [31]:
tsExecuteEngine.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^DJI" ')
Out[31]:
In [32]:
tsExecuteEngine.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^IXIC" ')
Out[32]:
In [33]:
tsExecuteEngine.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^GSPC" ')
Out[33]:
SW-WFA-Backtest 3¶
Description:
- Portfolio has all the trade strategies in Portfolio G
- Re-fit AI trading models every 3 months within a backtest and anchored within the backtest
- Fit datasets are anchored and duration increased by one year increments for five
consecutive years
- Every sliding window iteration, the training set duration is increased by 9 months and the validation set by 3 months, the test set is just shifted by one year.
- Training-set is anchored in outer slide window and also anchored in the re-fitting process during a backtest
- $200K/strategy
Backtest:
Type: Slide Window WFA Backtest
Iteration 1:
train: 1/1/1995 to 12/31/2000 validation: 1/1/2001 to 12/31/2002 Test Period: 1/1/2003 to 12/31/2003Iteration 2:
train: 1/1/1995 to 9/30/2001 (date range increased by nine-months) validation: 10/1/2001 to 12/31/2003 (date range increased by three-months and shifted by nine-months) Test Period: 1/1/2004 to 12/31/2004 (date range shifted by one-year)Iteration 3:
train: 1/1/1995 to 6/30/2002 (date range increased by nine-months) validation: 7/1/2002 to 12/31/2004 (date range increaed by three-months and shifted by nine-months) Test Period: 1/1/2005 to 12/31/2005 (date range shifted by one-year) ... ...Iteration Last:
train: 1/1/2000 to xxx validation: yyy to 12/31/2006 Test Period: 1/1/2007 to 12/31/2007
- Each iteration corresponds to a WFA Backtest, and in each WFA Backtest, AI Models are re-fitted every three months with the training and validation set for the WFA backtest anchored.
Portfolio Trade Assets¶
In [5]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [6]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "1980-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData_From_1980",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [7]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [8]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
"stop_lost" :
{ 'model' : "fix_perc",
'type' : "ta",
'obj_args' : { 'long_stop_los_per' : 5, 'short_stop_los_per' : 5 } }
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_wavenet_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 40, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #9
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #10
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 60 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #11
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_bb",
'actions_conf' : { "entry_exit" :
{ 'model' : "bb",
'type' : "ta",
'obj_args' : { 'window' : 13, 'window_dev' : 1 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #12
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_macd",
'actions_conf' : { "entry_exit" :
{ 'model' : "macd",
'type' : "ta",
'obj_args' : { 'window_slow' : 26, 'window_fast' : 12, 'window_sign' : 9,
'enter_long_macd_diff' : 0.5 , 'exi_long_macd_diff' : 0.5,
'enter_short_macd_diff' : -0.5, 'exi_short_macd_diff' : -0.5 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #13
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_rsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "rsi",
'type' : "ta",
'obj_args' : { 'rsi_enter_long' : 30.0, 'rsi_exit_long' : 30.0,
'rsi_enter_short' : 70.0, 'rsi_exit_short' : 70.0 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #14
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_srsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "srsi",
'type' : "ta",
'obj_args' : { 'window' : 14, 'smooth1' : 3, 'smooth2' : 3,
'srsi_enter_long' : 0.8, 'srsi_exit_long' : 0.8,
'srsi_enter_short' : 0.2, 'srsi_exit_short' : 0.2 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #15
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_adx",
'actions_conf' : { "entry_exit" :
{ 'model' : "adx",
'type' : "ta",
'obj_args' : { 'window' : 14, 'adx_enter_long_ge' : 14, 'adx_exit_long_lt' : 14,
'adx_enter_short_ge' : 14, 'adx_exit_short_lt' : 14 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #16
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_kama",
'actions_conf' : { "entry_exit" :
{ 'model' : "kama",
'type' : "ta",
'obj_args' : { 'window_fast' : 10, 'pow1_fast' : 2, 'pow2_fast' : 30,
'window_slow' : 10, 'pow1_slow' : 5, 'pow2_slow' : 30 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #17
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_tsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "tsi",
'type' : "ta",
'obj_args' : { 'window_slow' : 25, 'window_fast' : 5 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #18
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_forceindex",
'actions_conf' : { "entry_exit" :
{ 'model' : "forceindex",
'type' : "ta",
'obj_args' : { 'window' : 39 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #19
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_mfi",
'actions_conf' : { "entry_exit" :
{ 'model' : "mfi",
'type' : "ta",
'obj_args' : { 'window' : 14, 'long_slope_win' : 15, 'short_slope_win' : 15 },
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [9]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "SW-WFA-Backtest 3",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [10]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Training, Validation and Test Periods for Backtest¶
In [11]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "1994-07-01",
strBegDate = "1995-01-01",
strEndDate = "2000-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2000-07-01",
strBegDate = "2001-01-01",
strEndDate = "2002-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2002-07-01",
strBegDate = "2003-01-01",
strEndDate = "2003-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute Slide Window WFA Backtest¶
In [12]:
tsExecuteEngine = BacktestExecutionSlideWindowWFA( \
name = "SW-WFA-Backtest 3",
trade_asset_registered_item = portf_registered_obj,
wfa_slide_win_size_list = [ 3 ] , # list of int, associated with wfa_slide_win_period_list
wfa_slide_win_period_list = [ WFASlideWindowPeriod.Months ],
wfa_sw_scenario_obj = WFASlideWinScenarioAnchoredSlidingDatasets(\
fit_dataset = fit_datasets,
test_period = testing_period,
train_slide_win_size = 9, # increase train set 9 months every sw incr
val_slide_win_size = 3, # increase val set 3 months every sw incr
# and shift by 9 months
slide_win_period = WFASlideWindowPeriod.Months,
end_date_last_test_period = datetime(year=2007, month=12, day=31) ),
anchor_fit_dataset = True )
In [13]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [14]:
tsExecuteEngine.getSummaryMetrics().to_excel("SW-WFA-Backtest-3 backtest-metrics.xlsx")
DJI¶
In [15]:
tsExecuteEngine.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^DJI" ')
Out[15]:
Nasdaq¶
In [16]:
tsExecuteEngine.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^IXIC" ')
Out[16]:
S&P 500¶
In [17]:
tsExecuteEngine.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^GSPC" ')
Out[17]:
SW-WFA-Backtest 4¶
Description:
- Portfolio has all the trade strategies in Portfolio G and C
- Five Stop-Models are added utilized
- One model with fix values for stop-loss and trail-stop loss
- Four ATR models, each configured with different hyper parameters
- Every trade strategy in Portfolio C, and G, for each of their Enter/Exit Models is evaluated with each stop-model, including the composite Enter/Exit models
- The total number of permutations of all trade strategies for all securities, enter/exit models, stop models, and ensemble models is 633
- Re-fit AI trading models every 4 and 12 months within a backtest and not anchored within the backtest
- Training-set is not anchored in outer slide window and also not anchored in the re-fitting process during a backtest
- $200K/strategy
Backtest:
Type: Slide Window WFA Backtest
Iteration 1:
train: 1/1/1994 to 12/31/1998 validation: 1/1/1999 to 12/31/2000 Test Period: 1/1/2001 to 12/31/2001 ... ...Iteration Last:
train: 1/1/2003 to 12/31/2007 validation: 1/1/2008 to 12/31/2009 Test Period: 1/1/2010 to 12/31/2010
Portfolio Trade Assets¶
In [18]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [19]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "1980-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData_From_1980",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [21]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [22]:
tsConfDat_SuperSet = TradeStrategyConfigData()
# Set Config
tsConfDat_SuperSet.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "TA" ],
'ts_unique_tag' : "ts_macd_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "macd",
'type' : "ta",
'obj_args' : { 'window_slow' : 18, 'window_fast' : 7, 'window_sign' : 9,
'enter_long_macd_diff' : 0.5 , 'exi_long_macd_diff' : 0.5,
'enter_short_macd_diff' : -0.5, 'exi_short_macd_diff' : -0.5 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #2
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "TA" ],
'ts_unique_tag' : "ts_bb_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "bb",
'type' : "ta",
'obj_args' : { 'window' : 7, 'window_dev' : 1 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #3
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "TA" ],
'ts_unique_tag' : "ts_sma_1x",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #4
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "TA" ],
'ts_unique_tag' : "ts_macd_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "macd",
'type' : "ta",
'obj_args' : { 'window_slow' : 26, 'window_fast' : 12, 'window_sign' : 9,
'enter_long_macd_diff' : 0.5 , 'exi_long_macd_diff' : 0.5,
'enter_short_macd_diff' : -0.5, 'exi_short_macd_diff' : -0.5 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #5
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "TA" ],
'ts_unique_tag' : "ts_rsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "rsi",
'type' : "ta",
'obj_args' : { 'rsi_enter_long' : 30.0, 'rsi_exit_long' : 30.0,
'rsi_enter_short' : 70.0, 'rsi_exit_short' : 70.0 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #6
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "TA" ],
'ts_unique_tag' : "ts_srsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "srsi",
'type' : "ta",
'obj_args' : { 'window' : 14, 'smooth1' : 3, 'smooth2' : 3,
'srsi_enter_long' : 0.8, 'srsi_exit_long' : 0.8,
'srsi_enter_short' : 0.2, 'srsi_exit_short' : 0.2 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #7
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "TA" ],
'ts_unique_tag' : "ts_adx",
'actions_conf' : { "entry_exit" :
{ 'model' : "adx",
'type' : "ta",
'obj_args' : { 'window' : 14, 'adx_enter_long_ge' : 14, 'adx_exit_long_lt' : 14,
'adx_enter_short_ge' : 14, 'adx_exit_short_lt' : 14 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #8
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "TA" ],
'ts_unique_tag' : "ts_kama",
'actions_conf' : { "entry_exit" :
{ 'model' : "kama",
'type' : "ta",
'obj_args' : { 'window_fast' : 10, 'pow1_fast' : 2, 'pow2_fast' : 30,
'window_slow' : 10, 'pow1_slow' : 5, 'pow2_slow' : 30 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #9
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "TA" ],
'ts_unique_tag' : "ts_tsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "tsi",
'type' : "ta",
'obj_args' : { 'window_slow' : 25, 'window_fast' : 5 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #10
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "TA" ],
'ts_unique_tag' : "ts_forceindex",
'actions_conf' : { "entry_exit" :
{ 'model' : "forceindex",
'type' : "ta",
'obj_args' : { 'window' : 39 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #11
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "TA" ],
'ts_unique_tag' : "ts_mfi",
'actions_conf' : { "entry_exit" :
{ 'model' : "mfi",
'type' : "ta",
'obj_args' : { 'window' : 14, 'long_slope_win' : 15, 'short_slope_win' : 15 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #12
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "AI" ],
'ts_unique_tag' : "ts_gru_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #13
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "AI" ],
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #14
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "AI" ],
'ts_unique_tag' : "ts_conv1d_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #15
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "AI" ],
'ts_unique_tag' : "ts_conv1d_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #16
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "AI" ],
'ts_unique_tag' : "ts_wavenet",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #17
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "AI" ],
'ts_unique_tag' : "ts_wavenet_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #18
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "AI" ],
'ts_unique_tag' : "ts_wavenet_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "wavenet_lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #19
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "AI" ],
'ts_unique_tag' : "ts_gru_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 40, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #20
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "SMA" ],
'ts_unique_tag' : "ts_sma_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 20, 'win_slow' : 100 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #21
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "SMA" ],
'ts_unique_tag' : "ts_sma_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #22
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "SMA" ],
'ts_unique_tag' : "ts_sma_3",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 13, 'win_slow' : 65 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #23
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "SMA" ],
'ts_unique_tag' : "ts_sma_4",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 60 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #24
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "SMA" ],
'ts_unique_tag' : "ts_sma_5",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 75 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #25
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "SMA" ],
'ts_unique_tag' : "ts_sma_6",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 7, 'win_slow' : 35 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #26
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "SMA" ],
'ts_unique_tag' : "ts_sma_7",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 12, 'win_slow' : 40 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #27
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "SMA" ],
'ts_unique_tag' : "ts_sma_8",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 17, 'win_slow' : 80 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #28
tsConfDat_SuperSet.addTSConf(\
ts_conf_dict = {
'set_categories' : [ "SMA" ],
'ts_unique_tag' : "ts_sma_9",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 22, 'win_slow' : 90 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
Risk-Stop-Loss Models¶
In [23]:
# ***
# *** Configure Several Types of Stop Actions
# ***
# order_type = 1 ( Stop-Loss )
# order_type = 2 ( Trail-Stop-Loss )
# order_type = 3 ( Combine Stop-Loss & Trail-Stop-Loss )
#
# Stop Models :
# fixval { user-defined stop-loss-percent and/or trail-stop-loss-percent }
# atr { uses ATR model to dynamically determined the stop and trail stop loss percents }
# *** Stop-Loss Order Type
tsStopsConfDat_Stop_Loss = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 1 } }
} )
# 1) *** Combine Stop-Loss & Trail-Stop-Loss Order Type - Using Fixed Values
tsStopsConfDat_Stop_Loss_And_Trail_SL_1 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_1.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_1.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_fixval",
'actions_conf' : { "stop" :
{ 'model' : "fixval",
'type' : "ta",
'obj_args' : { 'stop_loss_per' : 0.002, 'trail_stop_loss_per' : 0.005 },
},
},
} )
# 2) ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_2 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_2.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_2.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_2",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.05,
'trail_stop_loss_mult' : 0.075,
},
},
},
} )
# 3) ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_3 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_3.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_3.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_3",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.01,
'trail_stop_loss_mult' : 0.03,
},
},
},
} )
# *4) **
tsStopsConfDat_Stop_Loss_And_Trail_SL_4 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_4.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_4.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_4",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.0075,
'trail_stop_loss_mult' : 0.0095,
},
},
},
} )
# 5) ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_5 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_5.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_5.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_5",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.0025,
'trail_stop_loss_mult' : 0.0050,
},
},
},
} )
stopsConfDatList = [ tsStopsConfDat_Stop_Loss_And_Trail_SL_1, tsStopsConfDat_Stop_Loss_And_Trail_SL_2,
tsStopsConfDat_Stop_Loss_And_Trail_SL_3, tsStopsConfDat_Stop_Loss_And_Trail_SL_4,
tsStopsConfDat_Stop_Loss_And_Trail_SL_5
]
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [24]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "SW-WFA-Backtest 4",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat_SuperSet,
stop_config_models = stopsConfDatList,
set_cats = [ "All", "TA", "AI", "SMA",
["TA", "AI"], ["TA", "SMA"], ["AI", "SMA"] ],
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [25]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Training, Validation, and Test Periods - Start Date Ranges¶
In [26]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "1993-07-01",
strBegDate = "1994-01-01",
strEndDate = "1998-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "1998-07-01",
strBegDate = "1999-01-01",
strEndDate = "2000-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2000-07-01",
strBegDate = "2001-01-01",
strEndDate = "2001-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute Backtest¶
In [27]:
tsExecuteEngine = BacktestExecutionSlideWindowWFA( \
name = "SW-WFA-Backtest 4",
trade_asset_registered_item = portf_registered_obj,
wfa_slide_win_size_list = [ 4, 12 ] , # list of int, associated with wfa_slide_win_period_list
wfa_slide_win_period_list = [ WFASlideWindowPeriod.Months, WFASlideWindowPeriod.Months ],
wfa_sw_scenario_obj = WFASlideWinScenarioSlidingDatasets(\
fit_dataset = fit_datasets,
test_period = testing_period,
slide_win_size = 1,
slide_win_period = WFASlideWindowPeriod.Years,
end_date_last_test_period = datetime(year=2010, month=12, day=31) ),
anchor_fit_dataset = False )
In [28]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Save to Excel¶
In [29]:
tsExecuteEngine.getSummaryMetrics().to_excel("SW-WFA-Backtest-4 backtest-metrics.xlsx")
Performance Metrics¶
DOW
In [30]:
tsExecuteEngine.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^DJI" ')
Out[30]:
Nasdaq
In [31]:
tsExecuteEngine.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^IXIC" ')
Out[31]:
S&P 500
In [32]:
tsExecuteEngine.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^GSPC" ')
Out[32]:
SW-WFA-Backtest 5¶
Portfolio K: Long and Multiple Configured SMA Trade Strategies for US Indixes, with Stop Models
Description:
1. Seven distinct SMA Trade Strategies for enter/exit a trade, plus an ensemble trade strategy of all seven SMA models 2. Eight different stop-models 1. Stop Loss order with specified stop-loss percent limit 2. Trail Stop Loss order with specified trail-stop-loss percent limit 3. Combine Stop-Loss & Trail-Stop-Loss order with specified percent limits 4-8. Five different Combine Stop-Loss & Trail-Stop-Loss models, configured with dynamic derived stop-loss and trail-stop-loss percent limits using Average True Range(ATR) algorithm 3. $2,000,000/strategy 3. No-Rebalance during the backtest 4. 100% capital invested 5. Portfolio investment using equal weights for each trade strategyBacktest:
1. Type: regular 2. Test Period: 1. Begin Date = 2001-01-01 2. End Date = 2020-12-31
Portfolio Trade Assets¶
In [33]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [34]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "1980-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData_From_1980",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [35]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [40]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 20, 'win_slow' : 100 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 50 } ,
'entry_policy_args' : { 'lastn_match' : 2, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_3",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 10, 'win_slow' : 57 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #4
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_4",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 60 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #5
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_5",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 75 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #6
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_6",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 7, 'win_slow' : 35 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #7
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_7",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 12, 'win_slow' : 40 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
Risk-Stop-Loss Models¶
In [41]:
# ***
# *** Configure Several Types of Stop Actions
# ***
# order_type = 1 ( Stop-Loss )
# order_type = 2 ( Trail-Stop-Loss )
# order_type = 3 ( Combine Stop-Loss & Trail-Stop-Loss )
#
# Stop Models :
# fixval { user-defined stop-loss-percent and/or trail-stop-loss-percent }
# atr { uses ATR model to dynamically determined the stop and trail stop loss percents }
# 1) *** Stop-Loss Order Type
tsStopsConfDat_Stop_Loss = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 1 } }
} )
tsStopsConfDat_Stop_Loss.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_fixval",
'actions_conf' : { "stop" :
{ 'model' : "fixval",
'type' : "ta",
'obj_args' : { 'stop_loss_per' : 0.002, 'trail_stop_loss_per' : 0.005 },
},
},
} )
# 2) *** Trail-Stop-Loss Order Type
tsStopsConfDat_Trail_Stop_Loss = TradeStrategyConfigData()
tsStopsConfDat_Trail_Stop_Loss.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 2} }
} )
tsStopsConfDat_Trail_Stop_Loss.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "tsl_fixval",
'actions_conf' : { "stop" :
{ 'model' : "fixval",
'type' : "ta",
'obj_args' : { 'stop_loss_per' : 0.002, 'trail_stop_loss_per' : 0.005 },
},
},
} )
# 3) *** Combine Stop-Loss & Trail-Stop-Loss Order Type - Using Fixed Values
tsStopsConfDat_Stop_Loss_And_Trail_SL_1 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_1.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_1.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_fixval",
'actions_conf' : { "stop" :
{ 'model' : "fixval",
'type' : "ta",
'obj_args' : { 'stop_loss_per' : 0.002, 'trail_stop_loss_per' : 0.005 },
},
},
} )
# 4) *** Combine Stop-Loss & Trail-Stop-Loss Order Type - Using Dynamic Values from ATR
tsStopsConfDat_Stop_Loss_And_Trail_SL_2 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_2.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_2.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_1",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 14,
'stop_loss_mult' : 0.5,
'trail_stop_loss_mult' : 1.0,
'max_stop_loss_per' : 0.05,
'min_stop_loss_per' : 0.002,
'max_trail_stop_loss_per' : 0.05,
'min_trail_stop_loss_per' : 0.005,
},
},
},
} )
# 5) ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_3 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_3.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_3.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_2",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.05,
'trail_stop_loss_mult' : 0.075,
},
},
},
} )
# 6) ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_4 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_4.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_4.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_3",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.01,
'trail_stop_loss_mult' : 0.03,
},
},
},
} )
# 7) ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_5 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_5.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_5.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_4",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.05,
'trail_stop_loss_mult' : 0.075,
'max_stop_loss_per' : 0.005,
'min_stop_loss_per' : 0.002,
'max_trail_stop_loss_per' : 0.005,
'min_trail_stop_loss_per' : 0.002,
},
},
},
} )
# 8) ***
tsStopsConfDat_Stop_Loss_And_Trail_SL_6 = TradeStrategyConfigData()
tsStopsConfDat_Stop_Loss_And_Trail_SL_6.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "stop" : { 'order_type' : 3 } }
} )
tsStopsConfDat_Stop_Loss_And_Trail_SL_6.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "sl_tsl_atr_5",
'actions_conf' : { "stop" :
{ 'model' : "atr",
'type' : "ta",
'obj_args' : { 'atr_mov_ave' : 7,
'stop_loss_mult' : 0.05,
'trail_stop_loss_mult' : 0.075,
'max_stop_loss_per' : 0.005,
'min_stop_loss_per' : 0.002,
'max_trail_stop_loss_per' : 0.01,
'min_trail_stop_loss_per' : 0.005,
},
},
},
} )
stopsConfDatList = [ tsStopsConfDat_Stop_Loss, tsStopsConfDat_Trail_Stop_Loss,
tsStopsConfDat_Stop_Loss_And_Trail_SL_1, tsStopsConfDat_Stop_Loss_And_Trail_SL_2,
tsStopsConfDat_Stop_Loss_And_Trail_SL_3, tsStopsConfDat_Stop_Loss_And_Trail_SL_4,
tsStopsConfDat_Stop_Loss_And_Trail_SL_5, tsStopsConfDat_Stop_Loss_And_Trail_SL_6
]
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [42]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "SW-WFA-Backtest 5",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
stop_config_models = stopsConfDatList,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
In [43]:
print("Number of configured Trade Strategies: ", len(portf_registered_obj.taRegisterItems) )
Training, Validation, and Test Periods¶
In [44]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "1997-07-01",
strBegDate = "1998-01-01",
strEndDate = "1999-12-31" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "1999-07-01",
strBegDate = "2000-01-01",
strEndDate = "2000-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2000-07-01",
strBegDate = "2001-01-01",
strEndDate = "2001-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute Backtest¶
In [45]:
tsExecuteEngine_sw = BacktestExecutionSlideWindowWFA( \
name = "SW-WFA-Backtest 5",
trade_asset_registered_item = portf_registered_obj,
wfa_slide_win_size_list = [ 20 ] , # No use here, no AI Strategies, just TA ones
wfa_slide_win_period_list = [ WFASlideWindowPeriod.Months ],
wfa_sw_scenario_obj = WFASlideWinScenarioSlidingDatasets(\
fit_dataset = fit_datasets, # no use in this backtest
test_period = testing_period,
slide_win_size = 1,
slide_win_period = WFASlideWindowPeriod.Years,
end_date_last_test_period = datetime(year=2020, month=12, day=31) ),
anchor_fit_dataset = False )
In [46]:
tsExecuteEngine_sw.evaluate()
Backtest Metrics¶
Save to Excel¶
In [51]:
tsExecuteEngine_sw.getSummaryMetrics().to_excel("SW-WFA-Backtest-5 backtest-metrics.xlsx")
Execution & Investment Metrics¶
DOW
In [53]:
tsExecuteEngine_sw.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^DJI" ')
Out[53]:
Nasdaq
In [54]:
tsExecuteEngine_sw.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^IXIC" ')
Out[54]:
S&P 500
In [55]:
tsExecuteEngine_sw.getSummaryMetrics()\
[ [ 'wfa_sw_iter_no', 'wfa_slide_win_size', 'wfa_slide_win_period',
'ta_reg_item_name', 'is_portfolio', 'trade_asset', 'order_type',
'enter_num_models', 'enter_model', 'enter_model_type',
'exit_num_models', 'exit_model', 'exit_model_type',
'stop_model', 'stop_model_type',
'Start Period', 'End Period',
'Cumulative Return ', 'CAGR%', 'Sharpe', 'Max Drawdown ', 'Best Day ', 'Worst Day '] ].\
query('trade_asset == "^GSPC" ')
Out[55]:
Reports¶
The reports currently supported are provided by quantstats, https://github.com/ranaroussi/quantstats.
- Guidelines:
- Run a backtest (Backtest, WFA Backtest )
- Select the Report to execute
- Select a trade strategy from the backtest for the report
- Optional, select a trade strategy from the backtest to use as a benchmark for the report
In [ ]:
from trade_engine.reports.BacktestReports import BacktestReports
WFA-Backtest for Reports¶
Description:
- Portfolio has a subset of the trade strategies in Portfolio G
Backtest:
1. Type: WFA Backtest
2. AI Models Fit Date Ranges
1. Training Period:
1. Begin Date = 2008-05-01
2. End Date = 2010-04-30
2. Validation Period:
1. Begin Date = 2010-05-01
2. End Date = 2010-12-31
4. Test Period:
1. Begin Date = 2011-01-01
2. End Date = 2012-12-31 Portfolio Trade Assets¶
In [210]:
portfolio_trade_assets = [ "^DJI", "^IXIC", "^GSPC" ]
Market Data Manager¶
In [211]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "1980-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData_From_1980",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [212]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Models¶
In [213]:
tsConfDat = TradeStrategyConfigData()
# Set Config
tsConfDat.tsSetConfig(\
ts_set_conf_dict =
{ 'set_actions_conf' :
{ "entry" : { 'votingPolicy' : 1, 'incNoInvestInVotingDec' : False,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1, 'cross_over_signal_for_trading' : False },
"exit" : { 'votingPolicy' : 1,
'minPerVotingLong' : 0.0, 'minPerVotingShort' : 0.0,
'lastn_match' : 1 }
}
} )
# TS #1
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_1",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 } },
},
} )
# TS #2
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_lstm",
'actions_conf' : { "entry_exit" :
{ 'model' : "lstm",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #3
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_conv1d_gru",
'actions_conf' : { "entry_exit" :
{ 'model' : "conv1d_gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 50, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 5,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #8
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_gru_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "gru",
'type' : "ai",
'obj_args' : { 'n_steps' : 40, 'no_neurons_lay1' : 20,
'no_neurons_lay2' : 20, 'pred_ahead_periods' : 3,
'ext_trade_dyn' : True, 're_eval_exit_dyn' : True,
'epochs' : 20 }
}
} } )
# TS #10
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma_2",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 15, 'win_slow' : 60 } ,
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #16
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_kama",
'actions_conf' : { "entry_exit" :
{ 'model' : "kama",
'type' : "ta",
'obj_args' : { 'window_fast' : 10, 'pow1_fast' : 2, 'pow2_fast' : 30,
'window_slow' : 10, 'pow1_slow' : 5, 'pow2_slow' : 30 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
# TS #17
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_tsi",
'actions_conf' : { "entry_exit" :
{ 'model' : "tsi",
'type' : "ta",
'obj_args' : { 'window_slow' : 25, 'window_fast' : 5 },
'entry_policy_args' : { 'lastn_match' : 1, 'cross_enter_signal' : False },
'exit_policy_args' : { 'lastn_match' : 1 },
},
} } )
Trade Strategies, Trade Investment Manager, and Trade Asset Register Items¶
In [214]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio For Reports",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1
)
Training, Validation and Test Periods for Backtest¶
In [215]:
training_period = DatasetDef(datasetType = DatasetType.Training,
strBegDateAIModels = "2007-11-01",
strBegDate = "2008-05-01",
strEndDate = "2010-04-30" )
validation_period = DatasetDef(datasetType = DatasetType.Validation,
strBegDateAIModels = "2009-11-01",
strBegDate = "2010-05-01",
strEndDate = "2010-12-31" )
testing_period = DatasetDef(datasetType = DatasetType.Testing,
strBegDateAIModels = "2010-07-01",
strBegDate = "2011-01-01",
strEndDate = "2012-12-31" )
fit_datasets = FitDatasetsDef( trainingDataset = training_period,
validationDataset = validation_period )
Execute WFA Backtest¶
In [216]:
tsExecuteEngine = BacktestExecutionWFA( \
name = "WFA Backtest for Reports",
test_period = testing_period,
fit_datasets_start = fit_datasets,
trade_asset_registered_item = portf_registered_obj,
wfa_slide_win_size = 4,
wfa_slide_win_period = WFASlideWindowPeriod.Months,
anchor_fit_dataset = False )
In [217]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
In [225]:
tsExecuteEngine.getExecutionMetrics().query('Is_Portfolio == "Yes"')
Out[225]:
Trade Strategies Execution & Investment Metrics¶
DOW Jones
In [218]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("DJI")', engine='python')
Out[218]:
Nasdaq
In [219]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("IXIC")', engine='python')
Out[219]:
S&P 500
In [220]:
tsExecuteEngine.getExecutionMetrics()[ ['TA_Reg_Name', 'Is_Portfolio', 'Num_Trades_Completed',
'Init_Capital', 'Final_Capital', 'Cum_Return', 'Backtest_Status']].\
query('TA_Reg_Name.str.contains("GSPC")', engine='python')
Out[220]:
Full Tearsheet Report for a Trade Strategy¶
Composite Nasdaq Trade Strategy vs Nasdaq¶
In [222]:
BacktestReports.createTearsheetReport(\
backtest = tsExecuteEngine,
ta_reg_item_name = "all_All_^IXIC",
ta_reg_item_benchmark_name = "Long-Only Trade Strategy for ^IXIC ", # optional
report_type = "full" )
Composite DJI Trade Strategy vs DJI¶
In [223]:
BacktestReports.createTearsheetReport(\
backtest = tsExecuteEngine,
ta_reg_item_name = "all_All_^DJI",
ta_reg_item_benchmark_name = "Long-Only Trade Strategy for ^DJI ", # optional
report_type = "full" )
Portfolio¶
In [226]:
BacktestReports.createTearsheetReport(\
backtest = tsExecuteEngine,
ta_reg_item_name = "Portfolio For Reports",
ta_reg_item_benchmark_name = None, # optional
report_type = "full" )
HTML Tearsheet Reports¶
In [227]:
BacktestReports.createHTMLTearsheetReport(\
backtest = tsExecuteEngine,
ta_reg_item_name = "all_All_^IXIC",
ta_reg_item_benchmark_name = "Long-Only Trade Strategy for ^IXIC ", # optional
title = 'Strategy Tearsheet',
download_filename = 'ts-tearsheet.html' )
Other¶
Trade Fee Policy¶
Description:
1. $2.0M Budget
2. 100% capital invested
3. Apply 1% fit per trade transaction
Trade Strategies:
1. One trade strategy: Nasdaq using Simple Moving Average(SMA) for entering/exiting a trade
Backtest:
1. Type: regular
2. Test Period:
1. Begin Date = 2001-01-01
2. End Date = 2005-12-31 Trade Asset¶
In [201]:
trade_assets = [ "^IXIC" ]
Market Data Manager¶
In [202]:
dirPathMarketData = '../Data/US_Market_Indixes'
supportedTradeAssets = [ "^DJI", "^IXIC", "^GSPC", "UDOW", "SDOW", "SPXL", "SPXS", "TQQQ", "SQQQ" ]
beg_date_str = "2000-01-01"
end_date_str = None
ts_marketdat_mng = TradeStrategyMarketDataMng(\
marketDataProviderName = "yahoofinance",
supportedTradeAssets = supportedTradeAssets,
stock_price_col_name = "Adj Close",
fileNameMarketData = "US_Indixes_And_ETFs_MrkData",
dirPathMarketData = dirPathMarketData,
beg_date_str = beg_date_str,
end_date_str = end_date_str,
forceCreateMarketData = False,
interval = "1d" )
Simulation Broker¶
In [203]:
brokerTradingObj = SimBrokerTrading()
Enter & Exit Model¶
In [204]:
tsConfDat = TradeStrategyConfigData()
# SMA JSON Configuration
#
# Model Configuration:
# win_fast : fast moving average
# win_slow : slow moving average
# Signals to enter and exit a trade based on the fast ma crossing the slow ma
tsConfDat.addTSConf(\
ts_conf_dict = {
'ts_unique_tag' : "ts_sma",
'actions_conf' : { "entry_exit" :
{ 'model' : "sma",
'type' : "ta",
'obj_args' : { 'win_fast' : 20, 'win_slow' : 100 } },
} } )
Trade Strategy, Trade Investment Manager, and Trade Asset Register Item¶
In [205]:
taRegItemGen = TradeAssetRegisterItemGenerator()
portf_registered_obj = taRegItemGen.createTARegItem(
name_ta_reg_item = "Portfolio Test Fee Policy",
# *** Trade Assets & Market Data
portfolio_trade_assets = portfolio_trade_assets,
ts_marketdat_mng = ts_marketdat_mng,
dirPathMarketData = dirPathMarketData,
market_data_period = MarketDataPeriod.Day,
# *** Broker
brokerTradingObj = brokerTradingObj,
# *** T.S. Permutations for each of the following sets of models
enter_exit_config_models = tsConfDat,
# *** Investment Manager Params
minPeriodsBetweenTrades = 1,
initInvestmentPerTradeStrategy = 2000000.0,
capPercentToInvestPerTrade = 100.0,
minInvestPercent = 5.0,
enter_trade_next_open_period = True,
min_account_balance = 1000.0,
portfRebalanceFreq = PortfolioRebalanceFrequency.Never,
portfRebalancePeriods = 1,
# *** Fee Policy Params per Trade
tradeFeeType = TradeFeeType.FixPercentPerTransactionValue,
percentPerTransaction = 0.01 # 1% per trade
)
Execute Backtest¶
In [207]:
tsExecuteEngine = BacktestExecution( \
name = "Backtest for Testing Trade Fee Policy - flat fee per transaction",
timePeriod = DatasetDef(datasetType = DatasetType.Testing,
strBegDate = "2001-01-01",
strEndDate = "2005-12-31" ),
trade_asset_registered_item = portf_registered_obj )
In [208]:
tsExecuteEngine.evaluate()
Backtest Metrics¶
Execution & Investment Metrics¶
In [209]:
tsExecuteEngine.getExecutionMetrics()
Out[209]: